AU2003202324B2 - Common protein surface shapes and uses therefor - Google Patents

Description

WO 03/066678 PCT/AU03/00137 1

TITLE

COMMON PROTEIN SURFACE SHAPES AND USES THEREFOR FIELD OF THE INVENTION THIS INVENTION relates to a method of determining common threedimensional structural features of proteins and use of representations of these common structures in molecular database searching, in molecular engineering and in designing focussed molecular libraries. More particularly, this invention relates to the identification and representation of protein surfaces such as p-tums, loops and contact surfaces and the determination of grid points describing surface charge, or the determination of common locations and orientations of amino acid side-chains, simplified as Coc-Cp vectors. These protein surfaces are typically involved in interactions with other molecules such as other proteins, nucleic acids, metal ions, antigens, drugs and toxins although without limitation thereto. This invention therefore provides common three-dimensional structural features that can be used to search molecular databases for the purposes of identifying molecules that match these common three dimensional structural features. The common three dimensional structural features can also be used to engineer de novo molecules or molecular libraries that have one or more common three dimensional structural features. Molecules and molecular libraries may be useful for the purposes of drug discovery.

BACKGROUND OF THE INVENTION The chemical diversity possible amongst the suspected 10180 possible drug-like molecules is immense, and a given combinatorial library can only hope to capture a tiny fraction of this diversity space. Molecular library design strategies use chemoinformatic techniques to select a diverse set of molecules for library synthesis. The molecular selection process involves the calculation of the chemical characteristics of each member of the library, using hundreds of chemical descriptors. It is therefore possible to derive a "diverse" library, where the molecules differ from each other as much as possible in descriptor space, or a "focussed" library, where the molecules are similar in descriptor space to a known active. With hundreds of potential descriptors it is difficult to know which descriptors are important or essential for describing biological activity. These WO 03/066678 PCT/AU03/00137 2 approaches consequently optimise libraries in the chemical universe but do not identify molecules that could modulate biological function.

It is becoming evident that the synthesis of large combinatorial libraries makes sense only if guided by sound library design principles. It is generally accepted that focussing libraries can lead to a 10-100 fold increase in the discovery of "hits"(i. e candidate or lead molecules).

A significant number of pharmaceutical targets involve the mimicking or inhibition of protein interactions with other molecules. With the rapid advance of the human genome project, it is likely that many more protein interaction targets will be identified.

Proteins are amino acid polymers that fold into a globular, structure. This globular structure, in general, has a hydrophobic interior. The structure of proteins is defined by the polymeric nature of the backbone and includes secondary structure elements such as helices, sheets, loops and turns. Whilst the description of protein structure by the nature of its polymeric backbone (its "skeleton") is useful for comparing one protein to another, it is not useful when describing the structural elements of various molecular recognition events of proteins. This is because molecular recognition is a surface phenomenon and proteins use large flat surface areas ranging from 1150 to 4660 A 2 comprising on average 211 atoms from 52 amino acid residues. These binding surfaces may be continuous (such as P-tums and loops), or discontinuous surfaces that comprise 1- 11 segments (where a segment is separated by at least 5 amino acid residues and can be from a different secondary structure) with an average of 5 segments per interface.

OBJECT OF THE INVENTION The present inventors have realized that by creating focussed libraries of compounds that mimic common structural' elements of protein surfaces, the likelihood of that library containing a molecule that mimics or inhibits a proteinmolecule interaction will be enhanced.

It is therefore an object of the invention to identify common elements of protein surfaces.

WO 03/066678 PCT/AU03/00137 3 It is also an object of the invention to provide a method to identify or to de novo engineer one or more molecules that match common elements of protein surfaces.

SUMMARY OF THE INVENTION The present invention is therefore broadly directed to the identification of common, protein surface elements as descriptors of protein surfaces for use in molecular design, engineering and screening.

In a first aspect, the invention provides a method of producing a description of a common three-dimensional protein surface shape including the steps of: identifying a three-dimensional surface shape of each of a plurality of proteins; and (ii) creating one or more descriptors wherein each said descriptor represents a common surface shape of two or more proteins of said plurality of proteins.

In one embodiment, the three three-dimensional surface shape is identified as respective amino acid side-chain locations and orientations of two or more amino acids of each said protein.

According to this embodiment, at step (ii) each said descriptor represents a common location and orientation of the respective amino acid side chains.

Preferably, each amino acid side chain used to produce said descriptor is simplified as a CQ-Cp vector.

In another embodiment, the three three-dimensional surface shape is identified as a surface chare distribution of each said protein.

According to this embodiment, at step (ii) each said descriptor represents a common charged surface region of two or more proteins of said plurality of proteins.

Preferably, each charged surface region is represented by at least four grid points.

According to the invention, said two or more amino acids form at least part of a structural feature of each of said two or more proteins.

Preferably, said structural feature is, or comprises, a p-turn, a loop or a contact surface.

WO 03/066678 PCT/AU03/00137 4 In a second aspect, the invention provides a method of identifying one or more molecules having a common three-dimensional protein surface shape, said method including the steps of: creating a query using one or more descriptors that each represent a common three-dimensional protein surface shape; and (ii) using said query to search a database and thereby identify one or more entries in said database that correspond to one or more molecules that each match said descriptor.

In one embodiment, at step the descriptor represents a common amino acid side-chain location and orientation of two or more amino acids of each of two or more proteins.

In another embodiment, at step the descriptor represents a common protein surface charge shape of two or more proteins.

In yet another embodiment, the query comprises: a descriptor that represents a common amino acid side-chain location and orientation of two or more amino acids of each of two or more proteins; and a descriptor that represents a common protein surface charge shape of said two or more proteins.

Suitably, according to the second aspect, said query is used to search a computer-searchable database comprising a plurality of entries.

Preferably, each amino acid side chain used to produce said descriptor is simplified as a Ca-Cp vector.

Preferably, for the purposes of database searching, Ca-Cp vectors and/or surface charge grid points are represented as a distance matrix.

In a third aspect, the invention provides a method of creating a library of molecules including the steps of: searching a database to identify one or more entries corresponding to one or more molecules that each match a common protein surface shape; and.

(ii) using at least one of the one or more molecules identified at step (i) to create a library of molecules.

WO 03/066678 PCT/AU03/00137 In a particular embodiment, this third-mentioned aspect includes the step of creating a library of molecules from the one or more molecules identified as step (ii).

Said library of molecules may be a "virtual" library or a synthetic chemical library.

In a fourth aspect, the invention provides a method of engineering one or more molecules including the steps of: creating one or more descriptors that each represent a common three-dimensional protein surface shape; and (ii) engineering one or more molecules that respectively comprise one or more structural features according to the or each descriptor in Throughout this specification, unless otherwise indicated, "comprise", "comprises" and "comprising" are used inclusively rather than exclusively, so that a stated integer or group of integers may include one or more other non-stated integers or groups of integers.

BRIEF DESCRIPTION OF THE FIGURES AND TABLES Figure 1. Distribution of c,,-ca 4 distances of all four residues segments that are not helical nor P-sheets and that are found in high resolution and nonhomologous structure in Protein Data Bank 3 1 Figure 2. a) Each P-turn is represented by four Ca-Cp vectors highlighted by the dark triangle, b) To aid visualization of the spatial arrangement of the turn after clustering, the four torsional angles 01, 02, 03 and 04 are used as approximation to the 24 distances. c) The four torsional angles are plotted as a vector from (01, 02) (represented by the symbol to (93, 94).

Figure 3. Vector plot of the seven clusters obtained from the kth nearest neighbor cluster and the filtered nearest centroid sorting algorithms. A threshold of 0.65 RMSD was used.

Figure 4. Vector plots of the eight clusters formed from the clustering algorithm and explicit division of cluster three into two clusters.

Figure 5. Vector plots of the nine clusters forned from the clustering algorithm and explicit division of cluster three into two clusters and inclusion of the average structure of type I' in the initial seed. The last graph represents the WO 03/066678 PCT/AU03/00137 6 conformations that were rejected and the first nine graphs represents the nine clusters.

Figure 6. The P-turns within each of the nine clusters were superimposed onto the cluster's mean structure.

Figure 7. Top view of the p-turns structures in cluster two superimposed onto its mean structure. The figure shows that the backbone structures can vary significantly even-though the c,-cp vectors are distributed uniformly.

Figure 8. Superimposition of the mean structures of the nine clusters. The superimposition is based on the three atoms cl ca2 and Ca3.

Figure 9. Vector plots of the p-turns in each of the nine types of p-turns defined by Hutchinson and Thornton 2 5 The order of the plots are: type I, II, IV, II', Vial, VIa2, VIII and VIb.

Figure 10. Number of neighboring loops versus its frequencies for various values of NEIGHBOR LIMIT.

Figure 11. Plot of number of peaks, number of peaks with greater than twenty neighbors and number of unique peaks with greater than twenty neighbors as a function of NEIGHBORLIMIT.

Figure 12. Filtered centroid sorting algorithm was used with various TOLERANCE to obtain the 39 clusters. The percentage representation, the intracluster RMSD, the intercluster RMSD and the ratio of the latter two are calculated and plotted as a function of TOLERANCE.

Figure 13. The loops are assigned to one of the 39 seeds using various TOLERANCEs in the filtered centroid sorting algorithm. The resulting loops in cluster one are superimposed and displayed in stereo view to show the effect of the choice of the tolerance value. Each line connects the position of the c~ atom in white and the position of the cp atom in gray.

TOLERANCE A) 0.3, B) 0.5, C) 0.7 AND D) 0.9.

Figure 14. Histogram of the number of loop in each of the 39 clusters.

Figure 15. Vector plots of the 39 clusters, with cluster number starting from one and counting across a row before going to the next row. The last cluster, cluster 40, represents all the loops that have not been clustered according to our filtered centroid sorting algorithm.

WO 03/066678 PCT/AU03/00137 7 Figure 16. Tree diagram of obtained from average linkage clustering of the 39 clusters.

Figure 17. An example of a bowtie. The function d(x,y) represents the euclidian distance between point x and pointy. (H=Head, T=Tail).

Figure 18. Algorithm for finding matching frequency of motifs.

Figure 19. Algorithm for finding peak motifs.

Figure 20. One-pass algorithm for clustering motifs.

Figure 21. The span of the one-pass algorithm.

Figure 22. The greedy algorithm for clustering motifs.

Figure 23. The span of the greedy algorithm for clustering motifs.

Figure 24. The combined one-pass and greedy algorithm for clustering motifs.

Figure 25. The greedy algorithm with sealevel for clustering motifs.

Figure 26. The span of the greedy algorithm with sealevel.

Figure 27. Adaptive sealevel applied in combination with the greedy algorithm.

Figure 28. The one-pass algorithm with RMSD tolerance.

Figure 29. Number of motifs verses size of motifs.

Figure 30. Highest matching frequency for each family tolerance.

Figure 31. Representative 4-motif C29.

Figure 32. Representative 5-motif Figure 33. Representative 6-motif C1.

Figure 34. Representative 7-motif Figure 35. Pseudo-code for finding the matching frequency of surface patches.

Figure 36. Pseudo-code for initial patch creation.

Figure 37. Pseudo-code for the creation of higher order N-patches.

Figure 38. The complete graph K 9 for a 9-patch.

Figure 39. a) Some scaffolds that match the common p-turns motifs, b) Some scaffold that match the common loop motifs, and c) A scaffold that match a common six-residue protein-protein interaction surface. The common motifs of the queries are shown in thicker lines in section a, b and c.

Table 1. RMSD matrix of the results obtained from filtered centroid sorting refinement of the 7 clusters formed from the fourth cycle of the kth-nearest neighbor clustering algorithm.

WO 03/066678 PCT/AU03/00137 8 Table 2. RMSD matrix of the eight clusters formed from clustering algorithm and explicit division of cluster three into two clusters.

Table 3. RMSD matrix of the nine clusters formed from clustering algorithm, explicit division of cluster three into two clusters and explicit inclusion of the mean of type I' in the initial seeds.

Table 4. Comparison of the unique peaks obtained using a NEIGHBORLIMIT of 0.3 with the unique peaks obtained using higher NEIGHOUR_LIMIT of 0.4, 0.5, 0.6 and 0.7. 'The unique-peak number obtained using NEIGHBOR LIMIT of 0.3 that is most similar to the unique peak obtained using higher NEIGHBOR_LIMIT. 2 The unique-peak number obtained using the higher NEIGHBORLIMIT that is shown in the header. 3 The RMSD value from the superimposition of the unique-peaks obtained using the NEIGHBOR LIMIT of 0.3 and obtained using the corresponding higher NEIGHBOR LIMIT.

Table 5. RMSD matrix of the 39 clusters. All the loops of cluster x (row x) are superimposed to the peak structures of cluster y (column y) and the resulting average RMSD obtained is shown in row x and column y. Since the average RMSD for row x, column y is very similar to that of row y, column x, the two values are averaged and place in row x and column y where x is greater or equal to y. The intracluster RMSD are highlighted by background shading. The average intercluster RMSD that is within 0.2 from the highest intracluster RMSD of 0.56 are also highlighted by background shading.

Table 6. Summary of results for 4-residue motifs.

Table 7. RMSD values for clustering of 4-residue motifs for initial families TOL 0.75 and inter TOL 0.5 with sealevel 0.25 times the peak.

Table 8. Summary of results for 5-residue motifs.

Table 9. RMSD values for clustering of 5-residue motifs for initial families TOL 0.75 and inter TOL 0.7 with sealevel 0.125 times the peak.

Table 10. Summary of results for 6-residue motifs.

Table 11. RMSD values for clustering of 6-residue motifs for initial families TOL 0.75 and inter TOL 0.7 with sealevel 0.125 times the peak.

Table 12. Summary of results for 7-residue motifs.

Table 13. RMSD values for clustering of 7-residue motifs for initial families TOL n '7 orl TrT n 0 o r;th QP1iPPvl 0 17P times the neak.

WO 03/066678 PCT/AU03/00137 9 Table 14. The secondary structure classifications as made by DSSP.

Table 15. The secondary structure of the original clusters for 4-residue motifs.

Table 16. The secondary structure of the original clusters for 5-residue motifs.

Table 17. The secondary structure of the original clusters for 6-residue motifs.

Table 18. The secondary structure of the original clusters for 7-residue motifs.

Table 19. The proportion of motifs not spanned by a single a-helix.

Table 20. The classification of the best 30 seeds of secondary structure not spanned by a single a-helix.

Table 21. Summary of results for the non-single a-helix seeds.

Table 22. RMSD values for clustering of 4-residue motifs for non-single-a-helix seeds greedy tolerance 0.5A with sealevel 0.25 times the peak matching frequency.

Table 23. The secondary structure of the non-single-a-helix clusters clusters for 4-residue motifs.

Table 24. The proportion of motifs in the non-single-a-helix clusters that do not contain a single a-helix.

Table 25. The secondary structure of the non-single-a-helix clusters for 7-residue motifs.

Table 26. The y, z) coordinate of the mean structure of the 5 th -least-common cluster of the p-turns, loops and surface motifs.

WO 03/066678 PCT/AU03/00137 DETAILED DESCRIPTION OF THE INVENTION The present invention describes the clustering of charged protein surface patches or regions, and the clustering of side chains of continuous and discontinuous surfaces of proteins into distinct motifs. These motifs are used as "descriptors" to design molecules that mimic specific, common protein shapes.

This is achieved by using these common protein motifs as a screen in a virtual screening of a virtual library and in de novo molecular design and engineering.

This approach results in the discovery of molecules that mimic common protein shapes. The synthesis of individual molecules or libraries of molecules that mimic protein shapes will result in the discovery of biologically active molecules that are capable of modulating protein function. In this respect these motifs or descriptors are used to select molecules from the vast chemical universe that match common protein shapes.

In order to define the common structural features of proteins used in molecular recognition, the present inventors have derived a new classification system to describe protein structure, based on the location and orientation of side chains, and based on the shape of the charged protein surface patches or regions, such as in protein-protein interaction regions. In particular the present inventors have focussed on defining the side chain arrangements of P-tums and loops, as these are primarily responsible for molecular recognition, as well as the side chain arrangements of contact surfaces.

Therefore, the present invention provides the identification of common protein motifs and the use of these as descriptors in molecular design. Libraries of molecules that mimic protein shapes should be valuable for the discovery of biologically active molecules using high throughput screening methodologies.

These libraries would form the foundation for the discovery of novel drugs.

As used herein, by "protein" is meant an amino acid polymer. Amino acids may be D- or L-amino acids, natural and non-natural amino acids as are well understood in the art. Chemically modified and derivatized amino acids are also contemplated according to the invention, as are well understood in the art.

A "peptide" is a protein having less than fifty (50) amino acids.

A "polypeptide" is a protein having fifty (50) or more amino acids.

WO 03/066678 PCT/AU03/00137 11 As used herein, a "protein surface shape" is any three-dimensional property or feature of a protein surface, such as may be described according to amino acid side chain location and orientation or by surface charge distribution.

In one particular embodiment, the property or feature is a threedimensional side-chain location and orientation of each of two or more amino acids of a protein.

In another particular embodiment, the property or feature is a threedimensional charge distribution of one or more surface regions of a protein.

Suitably, the protein surface shape is of, comprises or derived from, a structural feature of a protein. Such a structural feature may, for example, be a contact surface that interacts with another protein or other molecule such as a nucleic acid, nucleotide or nucleoside ATP or GTP) carbohydrate, glycoprotein, lipid, glycolipid or small organic molecule a drug or toxin) without limitation thereto. Therefore, for the purposes of exemplification, a domain may be a ligand-binding domain of a receptor, a DNA-binding domain of a transcription factor, an ATP-binding domain of a protein kinase, chaperonin or other protein folding and/or translocation enzyme, a receptor dimerization domain or other protein interaction domains such as SH2, SH3 and PDB domains, although the skilled person will appreciate that the present invention is not limited to these particular examples.

Structural features of proteins may include loops, p-tums or other contact surfaces, helical regions, extended regions and other protein domains.

Preferred structural features are in the form of loops, P-tums or other contact surfaces More preferred structural features are loops and contact surfaces.

As used herein, "contact surfaces" are protein surfaces having amino acid residues that contact or interact with another molecule, such as another protein.

An example of a contact surface is the ligand-binding surface of a cytokine receptor, although without limitation thereto.

Contact surfaces may be composed of one or more discontinuous and/or continuous surfaces.

WO 03/066678 WO 03/66678PCT/ATJO3/00137 12 By "discontinuous protein surface" is meant a protein surface wherein amino acid residues are non-contiguous or exist in discontinuous groups of contiguous amino acid residues.

In this regard, it will be appreciated that P-trns and loops are examples of a "continuous protein surface'. That is, a protein surface that comprises a contiguous sequence of amino acids.

According to the invention, it is preferred that the location and orientation in 3D space of each amino acid side-chain is simplified as a Cca-CI3 vector.

In one embodiment, a "descriptor" is a representation of common, or at least topographically related, amino acid side-chain locations and orientations in 3D space derived from two or more amino acids of each of two or more proteins.

Typically, a descriptor corresponds to a cluster of Cca-CP vectors obtained from two or more P-tums, loops, protein contact surfaces, helices or other structural features. Clusters are essentially groupings of 1-turns, loops or protein contact surfaces with common 3D topography. Clusters may be created by any algorithm that compares similarity and/or dissimilarity between constituent Cct-Cj3 vectors of f3-turns, loops or protein contact surfaces. Examples of clustering algorithms are provided in detail hereinafter.

In another embodiment, a "descriptor" is a representation of one or more common, three-dimensional distributions of charge across one or more surface regions of two or more proteins.

According to this embodiment, it is preferred that said descriptor represents four or more grid points.

Preferably, respective grid points are 0.2 to 2.0 angstrom apart in threedimensional space.

In particular embodiments, respective grid points may be 0.2, 0.5, 1.0, 1.2, or 2.0 angstrom apart in three-dimensional space.

It will also be appreciated that grid point dimensions may be modified within the ranges recited above according as desired. For example, protein surface regions that contribute significantly to protein-protein interaction may be represented by relatively tighter, less spaced-apart grid points. Conversely, protein surface regions that have less contribution to protein-protein interaction may be represented by relatively lesser, more spaced-apart grid points.

WO 03/066678 PCT/AU03/00137 13 It will further be appreciated that in particular embodiments, descriptors of common surface shape may be in the form of "average" surface shape, inclusive of "mean", "median" and "mode" surface shape.

Preferably, the common surface shape is a "mean" surface shape.

As already described hereinbefore, a side-chain location and orientation in 3D space, preferably simplified as a Ca-Cp vector, is required of at least two amino acids of each said p-turn, loop or contact surface.

In one embodiment, side-chain location and orientation of four P-turn or loop amino acids is required.

In another embodiment, side-chain location and orientation of at least three amino acids is required for a contact surface.

In particular embodiments, four, five, six or seven amino acid side-chain locations and orientations are required for a contact surface.

In some cases, descriptors are produced from protein structural information extracted from a source database such as the Protein Data Bank. In such cases, it is preferred that only non-homologous protein chains with relatively low homology no greater than 25%) to other proteins are used. This reduces biased sampling caused by the presence in the source database of multiple structures that are minor variants of each other.

In other cases, descriptors may be produced from protein structural information produced de novo, or from X-ray crystallographic or NMR determinations of protein 3D structure.

In light of the foregoing, it will be appreciated that the invention provides classification of proteins according to common amino acid side-chain locations and orientations to thereby produce a representation of common spatial elements of protein surfaces.

As will be described in more detail hereinafter, the present inventors to date have identified at least 9 p-turn, 39 loop and 240 protein contact surface shapes that occur with high frequency and may be useful in molecular database screening and library design.

For the purposes of database searching, a number of options are available for suitable representation of Ca-Cf vectors, whether as a database entry or as a query:- WO 03/066678 PCT/AU03/00137 14 as a distance matrix; as a dihedral angle formed between respective Ca-Cp vectors; as angles al and a2 formed between respective Ca-Cp vectors.

Explanations of these representations are provided in Lauri Bartlett 56 and International Publication WO 00/ 23474.

A preferred representation of Ca-Cp vectors is as a distance matrix.

It will also be appreciated that a preferred representation of surface charge grid points is as a distance matrix.

A computer-searchable database may be an existing database such as the Protein Data Bank, Cambridge Structural Database, Brookhaven Database or may be a database constructed de novo. For the purpose of database searching, entries may be in the form of representations of proteins, peptides or other organic molecules. It is preferred that entries in searchable protein databases are in the form of charged surface or Ca-C3 simplifications of constituent amino acid side chains. In cases where the searchable database comprises non-protein organic molecules (such as the Cambridge Structural Database), entries may typically be represented according to charge surface or 3D coordinate of particular atoms or groups of atoms (such as particular carbon, nitrogen and oxygen atoms, for example).

Suitably, a computer program is used for database searching.

Preferably, said computer program is the VECTRIX program, as described in International Publication WO 00/23474, It will therefore be appreciated that the database searching Aethod of the invention is capable of identifying one or more molecules, or portions of said molecules, that mimic common protein surface shapes. These molecules may then be used to construct virtual or synthetic chemical libraries that have been focussed by selecting molecules that are more likely to mimic the comunon protein surface shapes.

.So that the invention may be more properly understood and put into practical effect, the skilled person is directed to the following non-limiting examples.

WO 03/066678 PCT/AU03/00137

EXAMPLES

1 The clustering of 3-turns 1.1 Background Protein structure comprises stretches of secondary structure (helices or p-sheets) that are joined by turns, which enable a reversal in chain direction. These turns are normally positioned on the surfaces of proteins and allow the formation of the globular protein interior 1 P-turs 2 4 are more common than the tighter coiled yturns and the looser coiled a-turns and have been defined as four residue segments of polypeptides in which the distance between Cai and cai+3 is less than 7 A, and that the central residues are not helical 5 p-turns encompass 25% of residues in proteins 6 are important for protein and peptide function 279 and are an important driving force in protein folding 2 0 Consequently, there have been numerous studies on the design and development of P-turn mimetics'12-22 Despite the importance of side chain spatial arrangement in molecular recognition, the conformations of P-turns are currently classified in terms of the main chain dihedral angles, 4 and 3 ,5, 2325 Although this classification of p-turns has been extremely useful and has been used widely to design peptidomimetics, it makes very little functional sense. Each type of p-turn in the current classification can have two or more clusters of side chain spatial arrangement, and different types of p-turs can have the same side chain spatial arrangement.

There have been two reports on the classification of p-turns based on the arrangements of the side chains. Whilst the P descriptor of Ball et al 27 defines some global structural characteristics of the turns, it is clearly an oversimplification, as it only considers two of the possible four side chain positions and used only a small data set of 154 experimentally derived 3-turns.

Garland and Dean 282 9 have found common motifs for four subsets of.p-turns, ca atom doublets, ca atom triplets, c,-cp vectors doublets and c,-cp vectors triplets by hierarchical clustering the conformations of side chains. However, the clustering was not based on experimental data, but was based on all possible permutations of selecting doublets or triplets out of each of the eleven idealized p-tum types.

WO 03/066678 PCT/AU03/00137 16 Furthermore, it is not necessary to explicitly identify subsets of the p-turns for mimicry as have done by Garland and Dean 2 8 29 This is because it is possible to derive these subsets from the whole p-turn using more sophisticated searching algorithms 30 then considered by Garland and Dean.

1.2 Method 1.2.1 p-turns clustering 1.2.1.1 Extraction of p-turns from Protein Data Bank (PDB) A high resolution and non-redundant database of p-turns are required for the determination of common p-turn motifs that exist in proteins. To ensure high quality data, only high-resolution structures with a resolution of 2 A and an R factor 20% were extracted from the 1997 release of the Protein Data Bank 31 Furthermore, to eliminate the biased sampling in the PDB caused by the presence of multiple structures that are minor variations of a particular protein chain, only non-homologous protein chains with 25% homology with other protein chains were used. The distribution of the Cai-ca4 distances of the resulting 3984 fourresidue segments that are not helical nor p-sheets is plotted in Figure 1 and a major peak is observed at ca~-ca4 distances of 5.5 A. In 1973, Lewis 24 concluded that p-turns have cil-ca4 distance of 7 A based on the distribution of the cal-c,4 distances of only eight X-ray diffraction determined structures. To remove any possible biases caused by noisy data, the outliners of the major peak at 5.5 A were removed by eliminating the turns with Cal-Ca4 distance of less than 5 A or greater than 6.2 A, resulting in 2675 p-turns in the database.

1.2.1.2 Representation of data The present inventors motivations for clustering using c,-cp vectors are several fold. The c,-cp vector describes the initiation of the side chain geometry, and is well defined experimentally as it is anchored to the backbone. This is in contrast to the more flexible penultimate side chain atoms. Importantly, most mimetic strategies involve anchoring c,-cp bonds to a non peptidic scaffold, the extra atoms of the side chain providing a degree of flexibility in molecular recognition. We therefore consider that clustering according to ca-cp vectors is functionally significant, when the aim is to use the motifs identified to design ^la.,loo +hlbt mptirh Cmp.ifif" mntifk WO 03/066678 PCT/AU03/00137 17 Each of the 20 naturally-occurring amino acids, except for glycine, posses a Ca-cp vector due to the covalent bond between the central a carbon and the P carbon of the side chain. For p-tums that contains a glycine, the glycine residue was mutated to alanine to generate the required cx-cp vector. This was achieved by superimposing an ideal alanine structure onto the n, Ca and c' atoms of the glycine residue.

An important advance in database searching has been made by representing 3D structures in terms of the relationship between atoms located in distance space, rather than Cartesian space 3 0 32 A location in distance space is defined by distances between atoms, expressed in the form of a distance matrix.

Distance matrices are therefore coordinate independent, and comparisons between distance matrices can be made without restriction to a particular frame of reference, such as is required using Cartesian coordinates. It is important to emphasize that an arrangement of atoms and its mirror image are described by identical distance matrices. A root mean squared deviation (RMSD) can be used to alleviate this ambiguity. The four c,-cp vectors of each p-tur are represented by a distance matrix rather than a Cartesian coordinate system. Since there are four pairs of distances between each pair of ca-cp vectors (cat-ca2, cal-cP2, cpi-Ca2 and cpt-cp 2 and there are six possible pairs of ca-cp vectors 1-3, 1-4, 2-3, 2-4 and then 24 distances are required to represent the 3D topography of a Pturn. The distances between ci-cpi were not included because these bonded distances are relatively invariant between p-turns when compared to the nonbonded distances used.

1.2.1.3 kth-nearest neighbor The kth-nearest neighbor clustering algorithm 3334 employed here for clustering of p-turns is basically a simple-linkage clustering algorithm 35 in which every member is initially assigned to a different cluster and clusters are subsequently merged if the minimum distance between a member of a cluster and a member of another cluster is less than some threshold. The kt-nearest neighbor clustering algorithm 3 3 3 4 differs from simple-linkage clustering algorithm in that the distance between members is replaced by a dissimilarity measure defined below.

WO 03/066678 PCT/AU03/00137 18 dk(X) is defined as the Euclidian distance from observation x to the k nearest observation. Vk(x) is defined as the volume enclosed by the sphere, centering at observation x and having a radius of dk(x). The density at observation x, is defined as k vk(x) N where N is the total number of observations. The dissimilarity measure between observations xi and xj, D(xi, xj), can be calculated from the following definitions. First, xi and xj are said to be adjacent if the Euclidean distance between the two points is less than dk(xi) or dk(xj). If the observation xi and observation xj are not adjacent, then the dissimilarity measure, D(xi, xj), is set to infinity. Otherwise, D(xi, xj) is defined as the average of the inverse of the density, i.e. D(xi, xj)=(1/f(xi) 1/f(xj))/2.

Clustering should group together regions of high density separated by regions of low density. Effectively, by defining the dissimilarity measure as the inverse of the density, this algorithm first groups together adjacent points or clusters that have high-density.

The kth-nearest neighbor algorithm from the commercially available SAS/STAT program 36 was used to cluster the distance matrices representing the topography of the ca-c, vectors of the p-tumns. The option is called the smoothing parameter. A small value of produces jagged density estimates and large numbers of clusters, and a large value of produces smooth density estimates and fewer clusters. A value of two was used because only a rough estimate of clusters is required here. The clusters obtained here are used as initial seeds for the filtered nearest centroid-sorting algorithm described below.

1.2.1.4 Filtered nearest centroid sorting clustering algorithm The nearest centroid sorting clustering algorithm by Forgy 3738 requires a prior estimate of some initial seeds. The algorithm assigns each observation to the nearest initial seed to form temporary clusters. The seeds are then replaced by the means of the temporary clusters and the process is repeated until no further changes occur in the clusters. After the hnearest neighbor clustering of the 3turns, a modified form of the 'nearest centroid sorting algorithm 37 filtered nearest centroid sorting clustering algorithm was used to refine the clustering. This method superimposes observations in Cartesian coordinate space and thus removes the mirror image problem inherited from the distance matrix WO 03/066678 PCT/AU03/00137 19 representation in the k nearest mean clustering algorithm. The reasons for this two-stage clustering process are: 1) Hierarchical clustering based on RMSD could not be used in the first place because the number of observations is larger than the limit set by the SAS/STAT program 36 and 2) Faster and leaner approximation methods, such as nearest centroid sorting 3738 or K-means clustering algorithm 39 could not be used without prior estimate of initial seeds.

The filtered nearest centroid sorting algorithm is basically the same as the nearest centroid sorting algorithm except that if the minimum RMSD of a p-turn to the mean structures is above some definable threshold, then the turn is considered too remote and therefore not assigned to the temporary clusters. In latter iterations, these unassigned turns are superimposed onto the new mean structures of the new temporary clusters and if the minimum RMSD is below the threshold, then they are assigned to the new temporary cluster. The aim of this filtering is to remove the turns that are very different from the mean structures, so as not to bias the mean. Furthermore, 100% of the p-tums do not need to be clustered, only a major proportion is required. The filtered nearest centroidsorting algorithm was implemented in a program entitled "fncsacluster analysis.cpp".

1.2.2 Cluster Analysis 1.2.2.1 Vector plots of p-turns It is difficult to visualise the 24 distances that represent the topography of a p-turn. A vector plot is used to aid the visualization by approximating the 24 distances with four torsional angles 01, 02, 03 and 04 (see Figure 01 is defined as the torsional angle between cpl, c.i, ca2 and cp2. 02 is defined as the torsional angle between cz2, ca2, Ca3 and cp 3 63 is defined as the torsional angle between cp 3 c3, Cu 4 and cp4; and 04 is defined as the torsional angle between cpi, ca,, Ca4 and cp4. Since the distances between adjacent ca atoms in a p-tum are relatively constant due to the nature of the peptide bond, the four torsional angles should represent the essential conformational feature of a p-tur. The four torsional angles are plotted as a vector from (01, 02) (represented by the symbol to (03, 94). Effectively, this plot approximates the 24 distances of p-turns to four torsional angles (01, 02, 03 and 04), which are plotted as a vector on a 2D WO 03/066678 PCT/AU03/00137 graph. The torsional angles are periodic. A value of x is equivalent to x-360, x+360 and so on. To remove the graphing problem associated with the periodic nature of the torsional angles, each torsional value is transformed into a period that is closest to the torsional angles of the first p-turn.

1.2.2.2 Visualizing p-turn clusters Another method to visualize the clusters of p-turns is to superimpose the 3D structures of all the turns in a cluster. Superimposition is performed from the four c,-cp vectors of a p-turn to the four co-cp vectors of the mean structure of the cluster. For glycine, the c,-cp vector is obtained by superimposing a standard alanine residue to the n, ca and c' atoms of the backbone of the glycine residue.

The "fncsa_cluster_analysis.cpp" program outputs the coordinates of the superimposed structures in a multi-structure pdb file format which is visualised using the program InsightII of Molecular Simulation Inc.

There are a few steps in the calculation of the mean of a cluster in Cartesian coordinate space. Firstly, an initial mean structure for a cluster is set to be the first p-tum that does not have glycine or proline residue. Then each p-tur is superimposed to this temporary mean structure based on the coordinates of the c,-cp vectors. After the superimposition, a new temporary mean structure is computed by averaging the x, y and z coordinates. The latter two steps are repeated until successive mean structures differ by less than some arbitrary threshold.

1.2.2.3 Calculation of the RMSD matrix of all the clusters RMSD matrix is calculated to examine the performance of the clustering by assessing the dissimilarity within and between clusters. Each cluster is compared with every other cluster so that the row and column number of the matrix represents the cluster number. The value in a cell at row x and column y represents the mean RMSD when all the p-turns in cluster x is superimposed to the mean structure of cluster y. The diagonal of the matrix with row x and column x represents intra-cluster RMSD while the other cells represents intercluster RMSD. Values in row x and column y are not necessarily similar to values in row y and column x because the former represent the mean RMSD of WO 03/066678 PCT/AU03/00137 21 the p-turns in cluster x superimposed onto the mean structure of cluster y and the latter represent the mean RMSD of the p-turns in cluster y superimposed onto the mean structure of cluster x. However, the two numbers are very similar.

1.3 Results 1.3.1 Clustering The k t nearest neighbor cluster algorithm was used to cluster the 2675 Pturns in the database. The mean structure (seed) of each of the outputted 570 clusters was calculated by averaging each of the 24 distances representing the topography of p-turns. In the second cycle, kh nearest clustering was performed on these 570 seeds and 117 seeds were obtained. The third cycle of kI nearest clustering produced 25 seeds and the fourth cycle produced 7 seeds. Both the 7 and 25 seeds were examined in more detail prior to the selection of final p-tur clusters.

To determine a reasonable value for the threshold used in the filtered nearest centroid sorting algorithm, the seven seeds obtained from the kth nearest neighbor clustering were refined using filtered centroid sorting algorithm with four different threshold values, an RMSD of 0.6, 0.65, 0.7 and infinity (no threshold at all). The results show that the lower the threshold, the higher the percentage of p-turns that are rejected (not assigned to a cluster). The percentages of rejection for the four threshold values are 19%, 14%, 8% and 0% respectively.

The RMSD matrices of the results were calculated and the average of the mean inter-cluster RMSD are 1.05, 0.95, 1.03 and 1.15 respectively. Ideally one would like clusters to differ as much as possible, and hence have a high inter-cluster RMSD. The mean inter-cluster RMSD was lowered in going from a threshold of infinity to 0.7 and to 0.65, however it got higher in going from 0.65 to 0.6. The average of the mean intra-cluster RMSD are 0.36, 0.36, 0.40 and 0.44 respectively. In this instance, low intra-cluster RMSD are favored, therefore emphasizing that the observations in each cluster are similar. There were improvements in the intra-cluster RMSD in going from a threshold of infinity to 0.7 and from 0.7 to 0.65. However there was no improvement in going from a threshold of 0.65 to 0.6. As a compromise of the conflicting interest of WO 03/066678 PCT/AU03/00137 22 percentage rejection, inter-cluster RMSD and intra-cluster RMSD, a filter threshold of 0.65 was chosen.

To determine if the 25 seeds from the third cycle or the 7 seeds from the fourth cycle of the P nearest neighbor clustering best represent the side chain spatial arrangements of p-tums, both the results were subjected to the filtered centroid sorting algorithm followed by the calculation of the RMSD matrix. The RMSD matrix for the 7-clusters is shown in Table 1. The Clustering process aims to define clusters that have low intra-cluster RMSD separated by high intercluster RMSD. For the 25 clusters, the average of the mean intra-cluster

RMSD

is 0.31, the average of the mean inter-cluster RMSD is 1.11 and the maximum mean intra-cluster RMSD is 0.42. For the 7 clusters (Table the average of the mean intra-cluster RMSD is 0.36, the average of the mean inter-cluster RMSD is 0.95 and the maximum mean intra-cluster RMSD is slightly higher, 0.49. The results show that the clustering into the 7 clusters is not as good as the clustering into the 25 clusters, the intra-cluster RMSD was larger (0.36 compared to 0.31) and the inter-cluster was smaller (0.95 compared to 1.11). However, since this is not a drastic difference and the 7-clusters still give reasonable intra-cluster RMSD, the more tractable 7-clusters result was preferred over the result.

1.3.2 Refinement of the clustering Vector graphs, as described in the method section, were used to visualize the seven-cluster result (Figure The figure shows that all the clusters except for cluster three have a reasonable uniform distribution from a single mode.

Cluster three, however, seems to have two modes. One mode with 04 700 and the other mode with 04 700. Furthermore, the RMSD matrix in Table 1 shows that cluster three has the most varied intra-cluster RMSD of 0.49. To determine if cluster three should remain as one cluster or should be divided into two clusters, a practical step was used in which cluster three was divided into two clusters (one cluster with 04 70° and another cluster with 04 70°) and the new result assessed by comparison with the original result. The resulting eight clusters were refined once more using the filtered nearest centroid-sorting algorithm. The RMSD matrix and the vector plot for the new eight clusters were calculated and the results are shown in Table 2 and Figure 4 respectively. In dividing cluster WO 03/066678 PCT/AU03/00137 23 three into two clusters, the mean intra-cluster RMSD has not changed significantly (from 0.36 to 0.35), the maximum intra-cluster RMSD improved from 0.49 to 0.43, the minimum intra-cluster RMSD improved from 0.31 to 0.29, the mean inter-cluster has changed from 0.95 to 0.93 and finally the percentage of turns represented by the clusters remained the same at 86%. The vector plot in Figure 4 shows that the p-turns in each cluster distribute within a narrow range about a single mode. These results suggested that the eight clusters system is a better representation of p-turn motifs compared to the seven clusters system.

It was observed that type I' p-turs were not included in any of the eight clusters, they were rejected in the filtered nearest centroid sorting clustering because their RMSD with the mean of the eight clusters were more than the threshold of 0.65. This reflects a weakness in the k h nearest mean algorithm as it does not identify a seed near the low frequency type I' p-turns. Since Figure 9 indicates that there is a cluster near the type I' p-turns, the mean of the type I' was calculated and the result was included together with the other eight initial seeds for the filtered nearest centroid sorting clustering. The RMSD matrix and the vector plot for the new nine clusters were calculated and the results are shown in Table 3 and Figure 5 respectively. By the addition of the type I' average structure into the initial seeds, the mean intra-cluster RMSD has not changed significantly (from 0.35 to 0.36), the minimum and maximum intra-cluster RMSD remained the same, the mean inter-cluster worsen (from 1.25 to 1.1) and the percentage of p-turns classified improved from 86% to 90%. The vector plots in Figure 5 shows that the p-turns in each cluster distribute within a narrow range about a singie mode. These results suggest that the nine clusters system is a reasonable representation of P-turn motifs.

1.3.3 Mean structures The final nine-cluster result was also visualized by superimposing each Pturn in the clusters onto the clusters' mean structure (Figure The visual result is consistent with the mean intra-cluster RMSD value of each cluster in Table 3.

The cluster with the least amount of c,-cp vector spread (cluster 2, Figure 6) corresponds to the smallest mean intra-cluster RMSD. It is interesting to note that the backbone structure can vary significantly although the c-cp vectors are WO 03/066678 PCT/AU03/00137 24 uniform within a cluster. A top view of the most uniform cluster, cluster 2 for example, shows that different backbone conformations can have similar Ca-cp vector spatial arrangement (Figure In this instance, type I and type II (-turns are presenting the same c«-cp vector spatial arrangement. To appreciate the difference between the clusters, the mean structures of each cluster were superimposed based on the Cel, ca and C3a atoms. The result of this superimposition is displayed in Figure 8. The lowest inter-cluster RMSD (0.59) in Figure 3 is between cluster 2 and cluster 4. The result in Figure 8 also demonstrates that cluster 2 (red) and cluster 4 (green) are most similar and furthermore, provides a visual aid to understanding the meaning of an intercluster RMSD value of 0.59. The highest inter-cluster RMSD (2.38) exists between cluster 6 and 9 (Table The result in Figure 8 also demonstrates that cluster 6 (dark blue) and cluster 9 (grey) differ significantly.

The mean members of each cluster then become a query in a database searching strategy. This is described in more detail in Section 4.

2 Clustering of Loops of Proteins 2.1 Background Loops are defined as any continuous amino acid sequence that joins secondary structural elements (helices and sheets). Consequently, loops are a superset of (3-tums since there is no restriction on Cai-ca4 distances (as described above). Loops often play an important function as exemplified by their roles in ligand binding 40 DNA-binding 4 1 binding to protein toxin 42 forming enzyme active sites 43 binding of metal ions 44 binding of antigens by immunoglobulins 4 binding of mononucleotides 46 and binding of protein substrates by serine proteases 47 Identifying common loops motifs, then using these as queries in virtual screening of virtual library strategies will provide a novel and powerful strategy for the design and synthesis ofbioactive molecules.

2.2 Methods and Results 2.2.1 Extraction of loops from Protein Data Bank A database of loops was created by first extracting well refined (resolution of 2.0A and R-factor 20%) and non-homologous 25%) protein chains 4849 from the 1999 release of the Protein Data Bank 31 The program STRIDE 50 was then used to identify secondary structural elements (helices and sheets) of these chains.

WO 03/066678 PCT/AU03/00137 The linking regions, defined as the remaining residues that link these secondary structural elements or the protein terminus, were used for further analysis. The linking regions that consisted of four or more amino acid residues were divided into four residue segments, resulting in a total of 23650 four-residue loops. 319 of those loops were rejected because the distance between backbone atoms n, ca and c' was not appropriate 0.8 or 2 2.0 Each of the remaining 23331 loops was then simplified into four Ca-cp vectors (Figure 2a). Our motivations for clustering using ca-cp vectors are several fold. The ca-cp vector describes the initiation of the side chain geometry, and is well defined experimentally as it is anchored to the backbone. This is in contrast to the more flexible penultimate atoms in the side chain. Furthermore, most mimetic strategies involve anchoring c,-cp bonds to a non-peptidic scaffold, the extra atoms of the side chain provides a degree of flexibility in molecular recognition. We therefore consider that clustering according to ca-cp vectors is functionally significant 1 especially when the identified common motifs are used to direct peptidomimetic development.

2.2.2 Systematic identification of highly populated conformations (seeds) The present inventors have identified the appropriate seed points in order to cluster the loops of proteins. In this section the present inventors describe a process for identifying these seeds. By comparing each of the 23331 loops with all the other loops, all of the similar loops ("neighbors") having a RMSD value of less than a constant, NEIGHBOR_LIMIT, were identified and counted for each loop. A plot of the number of neighboring loops versus its frequency (the number of times this number of neighbors was found) using various NEIGHBORLIMIT is shown in Figure 10. For most values of NEIGHBORLIMIT, as the number of neighboring loops increases, the frequency decreases and as the number of neighboring loops decreases, the frequency increases. However, with large NEIGHBORLIMIT (0.6 and the frequency maximum is located between 200 to 600 neighboring loops instead of locating near the lower number of neighboring loops. This means that for large NEIGHBORLIMIT, it is more frequent to have 200 to 600 neighbors rather than 0-100 neighbors. The figure also shows the maximum number of neighboring loops for a particular NEIGHBOR LIMIT, thereby giving an indication to the similarity between the loops. For example, the maximum number of neighboring loops is 1763 with WO 03/066678 PCT/AU03/00137 26 NEIGHBOR_LIMIT of 0.7, 526 with NEIGHBORLIMIT of 0.4 and 146 with NEIGHBORLIMIT of 0.25.

Now that the number of neighboring loops has been determined for each loop, the next step was to identify (given a specific loop and its neighbors) the loop that has the largest number of neighbors (a peak). That is, the loop is marked as a peak if all the neighboring loops have lower or equal number of neighboring loops. The number of peaks for various NEIGHBOR_LIMIT is shown in Figure 11. The figure shows that the number of peaks varied from 7818 to 13 as the NEIGHBOR_LIMIT varied from 0.25 to 0.7. However, since peaks with the number of neighbors of less than an arbitrarily chosen low SEALEVEL of 20 are not interesting as they are not significantly populated, the plot also shows the number of peaks with greater than the SEALEVEL number of neighbors. The number of peaks with greater than the SEA_LEVEL of ranges from 56 to 3, decreasing with increasing

NEIGHBOR_LIMIT.

Since some peaks can be quite similar to each other, filtering was performed to identify a set of unique peaks to represent the data. Two peaks are considered not sufficiently unique if the fraction of shared loops between the two peaks exceeds an OVERLAPLIMIT value that was set to 20% of the total number of loops in the two peaks. From peaks with the largest number of neighbors to peaks with the lowest number of neighbors, the peaks were filtered out if they are not sufficiently unique when compared with all the previous chosen unique peaks. This definition, which is based on the fraction of overlap, is more discerning than a definition that is based on the RMSD distance between the average structures of the peak. Figure 1i aiso shows the number of unique peaks as a function of NEIGHBOR_LIMIT. For NEIGHBOR_LIMIT of 0.4 and above, all the peaks were unique. However, for NEIGHBORLIMIT of less than 0.4, not all the peaks were unique.

As to the choice of the value for NEIGHBORLIMIT, the general principle is that the larger the NEIGHBORLIMIT value, the more the data is "generalized". We seek to find the largest generalization without the loss of the number of unique peaks with greater than 20 neighbors (SEA_LEVEL Figure 11 shows that the number of unique peaks with greater than twenty neighbors ranges from 3 to 40, decreasing with increasing

NEIGHBOR_LIMIT

WO 03/066678 PCT/AU03/00137 27 value. When increasing NEIGHBOUR_LIMIT from 0.25 to 0.3, Figure 11 illustrates that there is no significant reduction in the number of unique peaks.

This contrasts to every other increase in NEIGHBOUR_LIMIT. Consequently a NEIGHBOURLIMIT of 0.3 retains the number of unique peaks whilst increasing the "generalization" of the data.

To examine the implication of the choice of 0.3 for NEIGHBOR_LIMIT, we determined whether the unique peaks obtained using a higher NEIGHBOR LIMIT are a subset of those obtained using the lower NEIGHBORLIMIT of 0.3. For example, if the unique peaks for the NEIGHBOURLIMIT of 0.4 are a subset (are similar to) the unique peaks found for the NEIGHBORLIMIT of 0.3 then we would expect to find structurally similar unique peaks in both datasets. To do this, we systematically pair each of the unique peaks obtained using a higher NEIGHBOR_LIMIT with the most structurally similar unique peak obtained using a NEIGHBOR_LIMIT of 0.3.

The results of pairing the 39 unique peaks obtained by using a NEIGHBORLIMIT of 0.3 with all the unique peaks obtained by using higher NEIGHBORLIMIT of 0.4, 0.5, 0.6 and 0.7 are shown in Table 4. The RMSD between the pairs of unique peaks for all comparisons ranged from 0.0 to 0.88.

There are a significant number of peaks between datasets that have an RMSD of less then 0.3. This illustrates that in going to a higher NEIGHBORLIMIT similar unique peaks are found when compared to the unique peaks identified from the 0.3 NEIGHBORLIMIT. This is particularly true for unique peaks with high frequency. As you go down the rows of the table, the frequency of the unique neak decreases. As can be seen for low frequency unique peaks the structural match between the datasets can sometimes be poor. We conclude that the choice of a NEIGHBOR_LIMIT of 0.3 is a reasonable compromise between "generalization" and accuracy (number of unique peaks) and the resulting unique motifs are used as seeds for further clustering.

2.2.3 Filtered centroid sorting clustering After the systematic identification of unique peaks with greater than twenty neighbors, the filtered nearest centroid sorting algorithm described in section 1.2.1.4 was utilized to refine the clustering. The 39 unique peaks with r i r.r. nr w a initinl qpdC for our filtered centroid WO 03/066678 PCT/AU03/00137 28 sorting algorithm using various TOLERANCE. The percentage of data clustered, the average intracluster-RMSD, the average intercluster RMSD and the ratio of the latter two are plotted as a function of the TOLERANCE in Figure 12. The figure shows that as the TOLERANCE increases, the percentage of data clustered and the intra-inter-RMSD ratio increases. At a TOLERANCE of 0.3, the percentage of data clustered, the average intracluster RMSD, the average intercluster RMSD and ratio of the latter two are 12%, 0.22, 0.95 and 0.11, respectively; at a TOLERANCE of 0.6 RMSD they are 71%, 0.42, 1.94 and 0.21, respectively; and at a TOLERANCE of 0.9, they are 100%, 0.51, 1.90 and 0.27, respectively. Therefore, there are opposing forces in the choice of tolerance.

Higher tolerance is favored because of the greater percentage of data clustered, but is disfavored because of the higher intra to inter cluster RMSD ratio. Lower tolerance is favored because of the lower intra to inter cluster RMSD ratio but is disfavored because of the lower percentage of data clustered. From a mimetics perspective, it is more important to have a reasonable intracluster similarity (intra cluster RMSD) than to have high percentage clustered. To aid in the choice of intracluster RMSD and indirectly the choice of TOLERANCE, a plot of all the loops in cluster one formulated using various tolerances are shown in Figure 13.

The figure shows that with an average intracluster RMSD of 0.47, which corresponds with a TOLERANCE of 0.7 and 89% of the data clustered, the members in the cluster are sufficiently similar for loop mimetic purpose. The numbers of loops in each of the 39 clusters obtained using a tolerance of 0.7 is plotted in Figure 14. The least populated cluster has 307 loops and the most populated cluster have 1048 loops.

2.2.4 Vector plots of loops The vector graphs, as described in 1.2.2.1, for all the 39 clusters (Figure shows that the loops within a cluster have reasonably similar conformations.

2.2.5 Calculation of the RMSD matrix of all the clusters An RMSD matrix is calculated to examine the performance of the clustering, by assessing the similarity within and dissimilarity between clusters (Section The RMSD matrix, showing the average intra- and interclusters RMSD for all the 39 clusters, is shown in Table 5. The average WO 03/066678 PCT/AU03/00137 29 intracluster RMSD of 0.47 and the vector graphs for all the clusters in Figure show that the loops within a cluster have similar conformations. The loops between clusters are dissimilar as the average intercluster RMSD is 1.91.

2.2.6 Average linkage clustering algorithm Based on the above mentioned RMSD matrix, we applied the average linkage clustering algorithm 365253 to determine the structural relationship between the 39 clusters. In the average linkage clustering algorithm, 3 6 5 2 5 3 each structure is initially assigned to its own cluster of size one. Subsequently, clusters are merged if the average distance between all the structures in the two clusters fall within some threshold. The resulting hierarchical tree, obtained by applying the average linkage-clustering algorithm on the 39 clusters, is shown in Figure 16.

All the cluster numbers used in this paper follow the order from left to right of this hierarchical tree.

2.2.7 Distinct clusters Whilst each cluster contains a unique set of loops (no loops are in more then one cluster), do the 39 clusters represent overlapping-variation from a continuous spread or do they represent distinct clusters that do not overlap in hyperspace? To answer this question, we defined that two clusters are 'distinct' if the most frequent eighty-percent of the data in one cluster does not overlap with the most frequent eighty-percent of the data in the other cluster. The overlaps were computed for each of the following thirty-two descriptors of the loop conformation. Each Ca-cp vector pair has four distances (Cac-Ca2, Cli-Cp2, Cpl-Ca2, cp2-cp2), so the six possible c,-cp vector pairs 1-3, i-4, 2- 5- or me four c,-cp vectors in a loop results in 24 distance descriptors. Furthermore, we also utilized all the possible six torsional angle descriptors between the four cacp vector and two torsional descriptors Ca2-Ca2-ca3-Ca4 and c 2 -c 2-c3-Cp4.

Distinctions were made based on each of those thirty-two descriptors.

The maximum and minimum values which delineate the most frequent eighty percent of each distribution was not computed using the mean plus and minus some standard deviation because some of the spreads were not always a 'Normal' distribution. The maximum and minimum was computed by first WO 03/066678 PCT/AU03/00137 'binning' the data with respect to each of the thirty-two descriptors mentioned earlier. Then, the bins for each descriptor were sorted based on the frequency, from most frequent to least frequent. As the program traverses down the bins, the maximum and minimum of the descriptor are stored. The traversal is stopped when the program has traversed through at least eighty percent of the data.

Consequently, the stored maximum and minimum represent the values that delineate the top eighty percentage of the distribution. This method of finding maximum and minimum works well with single peak distributions. For distributions with two or more peaks, the spread covered by the maximum and minimum are over-estimated. In such case, the determination of distinct is underestimated.

For a particular descriptor, if the maximum and minimum of cluster X overlap with those of cluster then cluster X and. cluster Y is considered to be overlapping. On the other hand, if the maximum and minimum of cluster X does not overlap with the maximum and minimum of cluster Y, then cluster X is considered to be distinct from cluster Y. This analysis shows that 737 out of a total of 741 combinations of clusters are distinct. Since the analysis was based on individual descriptors, those non-distinct clusters could be overlapping or distinct if the analysis was extended to include combinations of the descriptors.

These common protein loop motifs we have identified can now be used in database searching strategies to identify molecules that match the shape of these motifs. This is described in more detail in section 3 The Clustering of Protein Contact Surfaces 3.1 Backgrouna The protein contact surfaces are comprised of a continous sequence of amino acid residues as well as discontinuous sequences of amino acid residues.

In the previous two sections; we have clustered the continuous loops of protein binding sites. In this section, we describe the clustering of the side chains of protein contact surfaces.

3.2 Method 3.2.1 Definition of residues in protein-protein interfaces At least four criteria have been used in the literature to define residues that are involved in protein-protein interfaces. Two residues in two different chains WO 03/066678 PCT/AU03/00137 31 are considered to be in the protein-protein interface if their ca atoms are less than 9.0 A apart or any atoms in one residue is within 5 A of any atom of the other residue or the distance between any atom of one residue to any atom within the other residue is less than the sum of their corresponding van der Waals radii plus 0.5 A and the van der Waals energy between the residues is less than kcal/mol. The results are quite uniform between the four criteria 5 4 Criterion three is used here. Furthermore, when the number of residues in an interface is less than 10, the interface is rejected because there is a high probability that this protein-protein interaction is a result of crystal packing.

3.2.2 Non-redundant dataset Tsai 5 4 have scanned 2814 PDB entries from the September 1994 release of the PDB database 31 and found 1629 two-chain interfaces. Out of the 1629 twochain interfaces, Tsai et a1 5 4 have extracted 351 non-redundant families through the usage of structural comparison algorithm, measure of similarity and clustering of the structures into families (http://protein3 d.ncifcrf. gov/tsai/frame/dataset.html).

A program, p-p_interface.cpp was written to extract the coordinates of the c,-cp vectors of the residues in the protein-protein interface. For Glycine residues, the coordinates of the cp atom was obtained by superimposing the n, ca and c' atoms of an ideal alanine onto the n, c, and c' atoms of the glycine. Non- Glycine residues without ep atom coordinates were not included in the dataset.

The 350 non-redundant families gave rise to 700 protein chains that consist of up to 150 residues that form contact in a single chain. This results in a very large daIasei. For example, to identify common motifs from groups of 3-residues (this is the smallest size we would consider from a mimetic perspective) there would be as many as 700 x (350)=385910000residue comparisons. Most clustering algorithms are too computationally expensive for such a large problem.

Consequently, we have developed new approaches to tackle the clustering of protein contact surfaces.

3.2.3 Identification of seeds for clustering.

Identifying common structures in protein contact surfaces is a significant challenge. The first step in determining the matching frequency of a group of residues is to develop a method to compare them. The large size of the database WO 03/066678 PCT/AU03/00137 32 excludes the root mean squared deviation (RMSD) algorithm which is computationally expensive. A simpler method commonly used in chemical and biological situations is distance geometry comparisons.

The simplicity of this distance geometry method allows for rapid geometrical comparisons. However, distance geometry comparisons do not return a value of how closely related two geometries are (like an RMSD value) but instead return a match if their distances are the same within a certain tolerance.

The geometric relationship between two residues can be represented by four "bowtie" distances (Figure 17). The tolerance, TOL, represents the maximum allowed difference accepted between these distances to record a match. So two groups of residues and (Figure 17) are matched within TOL if and only if

D

1 TOL and D TOL and TOL and d(C,

TOL

where d(x,y) is the Euclidean distance between the points x and y in three dimensional space. Typically TOL e [0.2,1.0]A.

The extension to groups of more than 2 residues is simple. The general rule is that a bowtie must be formed between each two-residue combination within the group. A group of size N contains exactly bowties. To check to see if two groups of N residues are matched, every possible rotation in all) of the groups (or every bowtie-bowtie comparison) can be considered. For example, two groups of 3 residues and contain 6 possible residue matchups

{D,E,F)

{D,F,E}

{E,D,F}

WO 03/066678 PCT/AU03/00137 33

{E,F,D}

{F,D,E}

{F,E,D}

So these two 3-motifs are matched if and and or and and or and and or and and or and and or and and or where denotes that motifs {A,B}and are equal within tolerance.

3.2.4 Extracting matching frequency of motifs To extract the most common motifs from the dataset, each motif is compared against all others. The set of all motifs that match a particular motif, within a tolerance called family tolerance (TOL), is referred to as the family of that motif. The cardinality of that family is the matching frequency for that motif.

The common motifs with high matching frequencies are those that make good candidates for seed points.

Pseudo-code for the generic algorithm for determining the matching frequency for each motif is given in Figure 18. This algorithm generates all motifs and their bowtie distances and then exhaustively compares each against all in the data set.

Initially, 3-motifs were formed by simply considering every feasible combination of 3 residues within each chain. However, the set of feasible motifs excludes ones that contain any bowtie distance greater than 25A, as we are only interested in common surface patches of this size. Although a very large data set is produced, it is still manageable for this algorithm. The data set produced for 4- WO 03/066678 PCT/AU03/00137 34 motifs formed this way is too large. Using information from 3-motifs the size of the data set for 4-motifs can be greatly reduced. Every 4-motif is formed by 4 3motifs. For example, the group is formed by the 4 groups and If any of these motifs of 3 did not frequently occur in the database then there is no chance of the motif being a highly common motif either. This greatly reduces the size of the data set and is essential for the construction of higher order groups. The same method is used for higher order motifs.

3.2.5 Finding the peak matching frequencies Now that the number of matching frequency has been determined for each motif, the next step was. to identify the motifs with 'peak' number of matching frequency. A motif is marked as a peak if all the other motif within the family of the motif have lower or equal matching frequency. The algorithm for searching for peaks in the data set is given in Figure 19. Initially the algorithm tags every motif as a peak. Subsequently, for every motif, A, in the data set, if any motif within the family of the motif A have a lower matching frequency, it is tagged as non-peak.

3.2.6 Algorithms for Clustering Motifs The objective of clustering methods, in this section, is to retrieve all related motifs about some peak motifs, with the proviso that not all motifs in the data set need be clustered.

The simplest method for clustering motifs is a reverse scan of the data set as for the original matching algorithm. This method passes through the data set once accepting into the cluster every motif that is matched. This procedure is very similar to the PAM non-hierarchical method 55 This algorithm is called the one-pass algorithm and is outlined in Figure 20. The span of the one-pass algorithm for a landscape of motifs is illustrated in Figure 21. There is also an assumption that all hills are symmetric about their peak motif. The tolerance, OTOL, assumes that the width of each hill is identical. The entire hill is rarely collected because the range of the tolerance is constant throughout the algorithm.

There is a possibility that some motifs collected belong to a different hill.

An algorithm that in part overcomes these difficulties is the greedy algorithm (Figure 22). This algorithm is very similar to the single linkage I;or.h;onal m,-nrl Pfc h lteitp.r i initialised as a seed Doint. Motifs are added WO 03/066678 PCT/AU03/00137 to each cluster if they match any motif within the current cluster within tolerance.

The algorithm moves down each peak until no other motifs in the data set match any motif within the cluster. This span is illustrated in Figure 23. A flaw of the greedy algorithm is that it continues to collect motifs until no others exist that match those in the cluster (within GTOL). There is a danger of collecting motifs that belong to another peak (if GTOL is too large) or not collecting enough (if GTOL is too small). There is also no guarantee that the distribution of motifs down each hill is consistent or even. There is also a possibility that the algorithm may not halt until every motif in the database is collected if GTOL is overly large.

A simple method of overcoming this problem is to apply an additional tolerance to the greedy algorithm. The combined one-pass and greedy algorithm applies a one-pass tolerance, OTOL, to the greedy algorithm to limit its span. It applies the additional constraint that every motif in each cluster has to be within OTOL of the seed motif. This algorithm is outlined in Figure 24.

A tolerance, 'sealevel', with respect to the matching frequency (rather than its geometry) of a motif is also effective in restricting the span of the greedy algorithm. All motifs with matching frequency below the sea level are discarded from the data set. Figure 25 shows how a sea level is applied to the single linkage algorithm. An illustration of the possible span of the algorithm is given in Figure 26. However, selecting the correct sealevel is not always easy. If set too low it will have little effect in restricting the span of the algorithm. If set too high some peaks (and hills) will be excluded from the data set altogether. As an illustration, the first peak from the left in Figure 26 has been totally excluded from the data set and the third peak has been restricted more so than the second.

Unless all peaks are about the same height the application of a sea level will handicap some hills more than others.

To overcome this problem the sea level for each peak can be set adaptively as shown in Figure 27. This is a more appropriate method for restricting the span of the algorithm. Each sea level can be scaled according to the frequency of the peak motif.

All methods discussed in this section can be altered for superimposition of motifs onto the seed motif. Figure 28 shows how this adjustment is made to the PAM method. Although this new tolerance minimised the RMSD value for each ^f llo+ir ttip nptk for enc+ WO 03/066678 PCT/AU03/00137 36 seed. These algorithms are slightly more expensive than their distance geometry counterparts.

3.2.7 Secondary structure analysis The analysis of the secondary structure of each residue in the motifs was conducted using the DSSP (Dictionary of Protein Secondary Structure) software 4 This software has been utilized widely throughout the literature for the classification of protein shapes into formal secondary structure. Given the 3D coordinates of residues within a protein, DSSP classifies the shape based on its refined expert system. There are 8 different secondary structure classifications considered by DSSP. They are described in Table 14 together with the abbreviations that will be adopted in this paper. The classification 'no assignment' refers to shapes that do not fit any of the other classifications defined.

3.3 Clustering Results 3.3.1 Determining the seed points The first step to determine the seed points for clustering was to calculate the matching frequency of each 3-motif in the dataset. This was the highest order of motif that was computationally feasible. An exhaustive combination of 3motifs produced 9,215,424 motifs as opposed to 197,712,949 4-motifs..

Three family tolerances (TOL) of 0.25, 0.5 and 0.75 A were chosen based on a number of sample calculations. 0.25 A was the lowest tolerance that produced meaningful generalisations about structure in the dataset while 0.75 A was the highest tolerance that was computationally feasible, especially as the number of residues to be clustered increased.

Following the calculation of the matching frequencies for 3-motifs for all three TOL, the 4-motifs were constructed. As discussed previously in the method section, these motifs were constructed based on the common 3-motifs. At this point, we eliminated the uncommon motifs by removing all motifs that have a matching frequency below the matching frequency sealevel. For example, all 3motifs with less than 30 matching frequency were excluded before 4-motifs were created. This level was selected based on trends seen in the dataset A matching frequency sealevel of 30 is insignificant when compared against the highest matching frequency (434 for TOL 0.25A). We also tested a matching frequency sea levels of 20 when forming 4-motifs from the 3-motifs, and found that fl- ,aho fr.nicv sealevel had no effect on the frequency of the WO 03/066678 PCT/AU03/00137 37 seed points. Given the sealevels of 30 for 3-motif, 5 for 4-motif, 0 for higher order motifs, the highest matching frequency for each tolerance for each N-motif (3 N 7) is given in Figure 30. Figure 29 shows the number of N-motifs created. The number of motifs for larger tolerances is greater because the matching frequencies are higher and hence less of the dataset is excluded.

Following the calculation of the matching frequency for all motifs within the dataset, peak motif geometries can be extracted from the dataset (Figure 19).

From the dataset of peak motifs, the 30 motifs with the largest matching frequency were selected to be the seed points for the clustering stage. This value is kept constant for all motif sizes and tolerances. Although chosen arbitrarily, it almost always includes all unique motifs that have up .to half the matching frequency of the highest value for that dataset. At the same time, there are not too many seed points as to get significant overlapping in the clusters. When only seed points are selected, this overlapping only occurs when large tolerances are adopted for the clustering algorithms. In addition, a plot of a histogram of the dataset reveals that significant amount of the original data is covered by the most common unique motifs.

3.3.2 The clusters After the determination of the seed motifs, two clustering methods were finally adopted: one-pass algorithm (Figure 20) using the same tolerance as the initial family tolerance and with no sealevel applied. Greedy algorithm with adaptive sealevel proportional to the peak (or seed) matching frequency (Figure 27). A number of different greedy tolerances (GTOL) were applied in an attempt to achieve as large a range as possible for the tightness of the clusters.

Different adaptive sealevels of 0.125, 0.25, 0.5 and 0.75 of the frequency of the seed was trailed during clustering.

The success of each clustering algorithm is determined by three pieces of information: the size of the clusters, the intracluster RMSD and the intercluster RMSD. The aim is to cluster as many motifs as possible, though the resulting clusters should contain motifs that are similar (minimise intracluster RMSD), though each cluster must differ as much as possible (increase intercluster RMSD).

Summary table for the 4-motif, 5-motif, 6-motif and 7-motif are given is given in the Table 6, 8, 10 and 12, respectively. The three different family WO 03/066678 PCT/AU03/00137 38 tolerances considered, 0.25, 0.5 and 0.7, are given in the second column of each summary table. Information about the clustering for family tolerance 0.25 A are not presented for 6-motifs or larger because the dataset was too small to extract meaningful clusters. The summary tables give the sum of the size of the clusters and the sum of the number of unique motifs within those clusters. The difference between the sum of the size of the clusters and the sum of the unique motifs in clusters give the number of motifs that occur in more than one cluster. The intracluster and intercluster RMSD are the average for each algorithm.

Table 7, 9, 11 and 13 contain the intracluster RMSD, the average RMSD of each motif in the cluster superimposed onto the average motif for that cluster (along the main diagonal), and the intercluster RMSD, the average RMSD of all motifs against the mean motif of other cluster (entries off the main diagonal). At the top of tables, the size of each cluster is presented.

The representative clusters for each N-motif is selected based on a simple criteria: select the clustering technique producing the largest amount of data, while having small intracluster RMSD and large intercluster RMSD. A small intracluster RMSD is less than 0.5A and a large intercluster RMSD is greater than A. These values were determined based on a visualisation of the resulting clusters. In addition these clusters should be relatively distinct, that is, not too many motifs occurring in more that one cluster.

The summary of results for the clustering of 4-motifs is given in Table 6.

The parameters producing the largest set of clusters were the greedy method, family tolerance 0.75 A, algorithm tolerance 0.5 A and sealevel 0.125 of the seed matching freauencv. However the average intracluster RMSD of 0.67 A for this method is far too high. The next largest span of the dataset was achieved by the same method with sealevel 0.25 of the seed matching frequency. This method produced clusters with average intracluster RMSD of 0.51 which is much more acceptable. So the clusters produced by this method were selected as representative of the dataset for 4-motifs. The specific information about the selected 4-motif clusters is given in Table 7. The RMSD values on the main diagonal are much smaller than other values in the table. As described previously in Section 2.2.3, we anticipate using the filtered-centroid sorting algorithm to further refine the clusters identified. A picture of one of these clusters (C29) is E7 ii WO 03/066678 PCT/AU03/00137 39 The summary of results for 5- motifs is given in Table 8. The selected representative clusters in this case were the greedy algorithm, family tolerance 0.75, greedy tolerance 0.7 A and sealevel 0.125 of the seed matching frequency.

The specific information about the selected clusters for 5-residue motifs in given in Table 9. An example of a cluster of 5-residue motifs (C30) is given in Figure 32.

Average clustering results for 6- motifs are presented in Table 10. The selected method in this case was the greedy algorithm, family tolerance 0.75 A, greedy tolerance 0.7A and sealevel 0.125 times the peak matching frequency.

Specific cluster information about the methods are presented in Table 11. A representative cluster for the selected clustering algorithm CI is given in Figure 33.

The summary of results for 7- motifs is given in Table 12. The selected representative clusters in this case are the greedy algorithm, family tolerance 0.75A, greedy tolerance 0.9A and sealevel 0.125 of the seed matching frequency.

Information about the specific clusters is presented in Table 13. An example of a visualised cluster of 7- motifs (CIO) is given in Figure 34.

3.3.3 Secondary structure of the clusters An analysis of the secondary structure of the seed of each cluster was undertaken as described in the method section. The results show that all the residues of each seed were classified as a-helix except for the seed of cluster C2 of the 4-motifs where the four residues were classified as extended P3-strand.

There was a possibility, however, that this secondary structure classification of the seeds may not agree with the classification of the average motif of each cluster. Table 15, Table 16, Table 17 and Tablel8 give the distribution of secondary structure throughout each of the 4, 5, 6, and 7-residue motif cluster, respectively. The results in these tables confirm that the secondary structures of the seeds are almost always consistent with the secondary structures of the motifs within the cluster. The only possible exception to this is cluster C2 for the 4motifs, where only 56% of motifs have shared secondary structure with the seed.

Most of the others (all a-helical) have over 90% in agreement with the seed, for all sizes of motifs.

WO 03/066678 PCT/AU03/00137 Even if all the residues in the seeds or clusters are a-helical, the seeds or clusters do not necessarily belong to a single a-helix because the residues flanking between the residues in the motif may not be a-helical. Due to possible uncertainty with the DSSP classification, an a-helix is considered broken when flanked by two or more consecutive non a-helical residues. Table 19 records the proportion of motifs in each cluster that are not single a-helix. Except for cluster C2 of the 4- motifs (of which 99% are non-helical), almost all other motifs are single a-helix.

3.3.4 Non single a-helical clusters Given that almost all the above mentioned clusters are part of a single ahelix, we proceeded with extracting non-single a-helix clusters. The first step was to find the highest matching frequency seeds that were not single a-helix.

The secondary structure of these seeds is given in Table 20. The results show that as the size of the motif increases, the secondary structure within each motif were more uniform.

Table 21 contains a summary of the clusters retrieved using a variety of methods with the new seeds. The most successful clustering methods of the previous analysis were adopted as starting points for this analysis. These results show that there is a strong trend towards clusters becoming more distinct as the size of the motif increases. For 4-residue motifs, there is significant sharing of motifs between clusters with many having quite a large intracluster RMSD.

However, as the size of the motifs become larger, significantly larger tolerances can be adopted without altering the composition of the resulting clusters. The greedy tolerance could be extended to 1.1A for 6-residue motifs with the intracluster RMSD remaining very low. This is in contrast to the highest tolerance of 0.7A that could be adopted for the original set of seeds.

The criteria for the selection of representative clusters for each size of motif was the same as for the previous clustering study: to select the largest clusters possible, while keeping the intracluster RMSD low (less than approximately 0.5A) and the intercluster RMSD high (greater than 2.0 A).

RMSD comparisons between these resulting clusters for 4-residue motifs, is presented in Table 22.

WO 03/066678 PCT/AU03/00137 41 Table 23 presents the distribution of different types of secondary structures for the new 4-residue motif clusters. Even though the seeds of each of these clusters are classified as 'not a single a-helix', a large proportion of motifs within three clusters are a-helix. Table 24 presents the proportion of motifs in each of the new clusters that are not a single a-helix. Less than 20% of motifs of cluster C5, C17 and C25 are not a single a-helix and more than 90% of residues of these clusters are classified as a-helix. Clusters that have a high proportion of motifs with no assignment to formal secondary structure, such as C2, C3, C4, C7, C12 and C16 have relatively small size (the largest has just 26 members) and relatively low intracluster RMSD. Within this group, the average intracluster RMSD is 0.28A as opposed to the average intracluster RMSD of 0.52A for the entire set.

This relationship remains consistent for larger sizes of motifs too. Table presents the distribution of secondary structure classifications for 7-residue motifs. Only clusters C1, C2 and C3 have motifs whose component residues are not entirely classified 'no assignment'. These three clusters have average intracluster RMSD of 0.41 A which is much larger than the average intracluster RMSD for the entire set of 0.26 A. This correlation suggests these clusters that contain motifs with no formal secondary structure assignment, have shape that is highly unique to any other motifs in the dataset.

4 Clustering of surface patches The basic algorithm for the clustering of surface patches is similar to that for clustering discontinuous protein surfaces. Firstly, snapshots of the protein surface, called patch motifs, is generated. A paich motif conmaining grid points is referred to as an N-patch. The smallest patch size that will be considered in this study is the 3-patch. The algorithm for the construction of 3-patches is given in Figure 36.

Figure 35 describe the algorithm for determining the matching frequency of each patch motif. Again, an RMSD superimposition of these motifs is too computationally expensive because of the size and the number of patches. A distance matrix is constructed so that the shape of the patches can be easily compared. This distance matrix is calculated for an N-patch by creating a complete graph KN whose vertices are the grid points and edges are weighted by WO 03/066678 PCT/AU03/00137 42 the Euclidian distance between each pair of vertices. An illustration of a complete graph K 9 constructed for a 9-patch is given in Figure 38.

A total of N! comparisons are required to determine if two N-patches are equal in shape. Every vertex-vertex match-up needs to be considered. This is a very expensive calculation computationally. If N=4, for example, 24 orientations with 144 edge distance comparisons, need to be attempted in order to determine if two patches do not have matching geometric structure. If N=9, the number of orientations becomes 362880 with 13 063 680 distance comparisons! A number of improvements need to be made to make this problem feasible.

A number of quick and simple comparisons can be made between pairs of patches before an exhaustive check should take place. This is to improve the computational feasibility of the problem. The first is to compare the charges of each grid point before comparing distances. If the charges don't match within the charge tolerance, there is no need to check if the edge distances match. This is a much easier calculation. Another is to check if the longest and shortest edge distances of the pair of patches match. If they don't then the geometric structure of the patches is different. Finally, there may be some distances of the distance matrix that remain constant for all patches because of the way the grid was originally constructed. This may remove the need to consider certain orientations of patches when being compared.

The N+l-patches will be constructed based on N-patches in similar manner to the higher order motif build-up procedure for the discontinuous surfaces described earlier. Patches that are less than a defined matching frequency sealevel will be removed from the dataset. This reduces the number of redundant higher order patches that will be created. The algorithm for the creation of higher order patches is given in Figure 37. In this algorithm, each new N-patch is created from (N-1)-patches.

Once the matching frequency is determined, seeds and clusters can be obtained as described in the previous sections.

5 Scaffolds As described above, the present inventors have clustered the side chain positions of 3-tums, loops and protein contact surfaces. This has resulted in the identification of 9, 39 and 240 highly populated motifs for P-turns, loops and WO 03/066678 PCT/AU03/00137 43 protein contact surfaces, respectively. As an example, the coordinate of the 5 t least popular cluster of all these motifs are given in Table 26. These motifs define common spatial elements of protein surfaces. Our objective is to use these motifs to design libraries of molecules. Consequently, these motifs are used as biological descriptors in library design, and the resulting libraries will mimic common protein shapes. In high throughput screening, such libraries will be a valuable resource for the development of new lead compounds.

To this end, the present inventors have used a subset of the motifs and have screened the virtual library of molecules derived from the Cambridge Structural Database to identify molecules that match the spatial elements of the motifs. Our in house virtual-screening of virtual-library program, VECTRIX, was used to search the database. Figure 39a shows some of the scaffolds that match the P-turn conformations, Figure 39b shows some of the scaffolds that match the common loop conformations, and Figure 39c shows a scaffold that match a common six-residues protein-protein interaction surface.

As illustrated in Figure 39, molecules are identified that match the shape of the common motifs. This information will lead to the design of molecules that match common protein shapes.

Throughout this specification, the aim has been to describe the preferred embodiments of the invention without limiting the invention to any one embodiment or specific collection of features. Various changes and modifications may be made to the embodiments described and illustrated herein without departing from the broad spirit and scope of the invention.

All computer programs, algorithms, patent and scientific literature referred to in this specification are incorporated herein by reference in their entirety.

WO 03/066678 WO 03/66678PCT/AU03/00137 44

REFERENCES

1) Rose, Young, W.B. Nature 1983, 304, 654-657.

2) Rose, Gierasch, Smith, J.A. A4dv. Protein. Chem. 1985, 37, 1-109.

3) Wilmot, Thornton, J.M. Protein engineering 1990, 3 479- 493.

4) Kabsch, Sander, C. Biopolymers 1983, 22,2577-2637.

Richardson, IS. Adv.Protein. Chem. 1981, 34, 167-339.

6) Wilmot, Thornton, J.M. JMol.Biol 1988, 203, 22 1-232.

7) Freidinger, Veber, Perlow, Brooks, J.R. Science 1980, 210, 656-658.

8) Li, Lee, Lee, Yoon, CIJ; Baik, Lim.S.K.

Eur.J.Biochem. 1999, 265, 430-440.

9) Andrianov, A.M. Molecular Biology 1999, 33, 534-538& Smith, Pease, L.O. CRC Crit. Rev. Biochem. 1980, 8, 3 15-399.

1 5 11) Mutter, M. TIBS 1988, 13, 260-265.

12) Ball, Alewood, P.F. Journal of Molecular Recognition 1990, 3, 55-56.

13) Tran, Treutlein, Burgess, A.W. JComputChem. 2001, 22 1010-1025.

14) Douglas, Mulholland, Walker, Guthie, Elmore, D.T.; Murphy, R.F. Biochem.Soc. Tranis. 1988, 16, 175-176.

Li, Burgess, K. Tetrahedron Lett. 1999, 40, 6527-6530.

16) Halab, Lubell, W.D. Journal of organic chemistry 1999, 64, 33 12- 332 1.

WO 03/066678 WO 03/66678PCT/AU03/00137 17) Terrett, N. Drug Discovery Today 1999, 4, 141-141.

18) Rosenquist, Souers, Virgilio, Schurer, Eliman, J.A.

Abstracts ofpapers of the American chemical society 1999 1999, 217, 212 19) Gardner, Liang, Gellman, S.H. JAm. Chem. Soc. 1999, 121, 1806-1816.

Mer, Kellenberger, Lefevre, J.F. J.MolBio 1998, 281, 235- 240.

21) Lombardi, D'Auria, Maglio, Nastri, Quartara, L.; Pedone, Pavone, V. JAm. Chem.Soc. 1998, 120, 5879-5886.

22) Fink, Kym, Katzenellenbogen, J.A. J.Am.Chem.Soc. 1998, 120, 4334-4344.

23) Venkatachalani, C.M. Biopolymers 1968, 6, 1425-1436.

24) Lewis, Momnany, Scheraga, H.A. Biochim Biophys Aceta 1973, 303, 211-229.

Hutchinson, Thornton, J.M. Protein Science 1994, 3, 22-7-22 16.

26) Ball, Hughes, Alewood, PEF.; Andrews, P.R. Tetrahedron 1993, 49, 3467-3478.

27) Ball, Andrews, Alewood, Hughes, R.A. Federation of European Biochemical Societies 9 9, i-3 28) Garland, S. Dean, P.M. J. Comput. -Aided Mot. Design 1999, 13, 469- 483.

29) Garland, Dean, P.M. J. Comput. -Aided Mot. Design 1999, 13, 485- 498.

Ho, Marshall, G.R. J Comput. -Aided MoL.Design 1993, 185, 3-22.

WO 03/066678 WO 03/66678PCT/AU03/00137 46 31) Berstein, Koetzle, Williams, Edgar, Meyer, Brice, Kennard, Shimanouchi, Tasumi, M. JMol.Biol. 1977, 112, 53 5-542.

32) Jakes, Willett, P. Journal of Molecular Graphics 1986, 4,12 33) Wong, Lane, T. J.R.Statist.Soc.B 1983, 4S,362-368.

34) Wong, Lane, T. JRkStatist. Soc. B 1983, 4S, 3 62-3 68.

Tsai, Lin, Wolfson, Nussinov, R. Protein Science 1997, 6, 53-64.

36) AnonyrnousSAS/STAT User's guide, Volume 1, ANOVA-FREQ, Version 6; 1999; 37) Anderberg, M.R.Cluster analysis-for applications; Academic Press: New York and London, 1973; 38) Forgy, E.W. Biometrics 1965, 21, 768 39) MacQueen, J.B. Proc. Symp. Math. Statist. and Probability 1967, 1, 281- 297.

Joseph, Petsko, Karplus, M. Science 1990, 249, 1425-1428.

41) Jones, van Heyningen, Berman, Thornton, J.M. JMMolBiol.

1999, 287, 877-896.

42) Wu, Dean, D.H. J. lvfol.Biol. 1996, 255, 628-640.

43) Wiodawer, Miller, Jakoiski, Sathyanarayana, B.K.; Baldwin, Weber, Selk, Clawson, Schneider, Kent, S.B.H.

Science 1989, 245, 616-621.

44) Lu, Valentine, J.S. Curr.Opin.Struct.Biol. 1997, 7, 495-500.

Bajorath, Sheriff, S. Proteins 2001, 24, 152-157.

WO 03/066678 WO 03/66678PCT/AU03/00137 47 46) Kinoshita, Sadanamni, Kidera, G6, N. Protein engineering 1999, 12, 11-14.

47) Perona, Craik, C.S. Protein Science 1995, 4, 337-360.

48) Hobobm, Scharf, Schneider, Sander, C. Protein Science 1992, 1, 409-417.

49) Hobohm, Sander, C. Protein Science 1994, 3, 522 Frishman, Argos, P. Proteins: structure, function and genetics 1995, 23, 566-579.

51) Damewood, J.R.In Lipkowitz, Boyd, Eds.; VCJIPublishers: New York, 1996; pp 1-79.

52) Sokal, Michener, C.D. University of Kansas Science Bulletin 1958, 38, 1409-1438.

53)1 Manly, B.F.J.Multivariate statistical method, A primer; Chapman Hall: London, 1994; 54) Tsai, Lin, Wolfson, Nussinov, R. J.Mol.Biol. 1996, 260, 604-620.

Kaufman, Rousseeuw, John Wiley and Solis Publisher: New York, 1990.

56) Lauri Bartlett Comp. A id. Mol. Des. 1994, 8 WO 03/066678 PCT/AU03/00137 Table 1.

Clusterl1 1 2 3 4 5 6 7 0.32 0.70 0.85 1.12 0.72 0.84 1.40 0.70 0.31 0.75 0.64 0.66 1.09 0.95 0.78 0.65 0.49 0.81 0.72 0.93 1.05 1.10 0.60 0.86 0.38 0.81 1.46 0.67 0.71 0.63 0.79 0.82 0.35 1.14 0.90 0.84 1.09 0.99 1.47 1.15 0.31 1.70 1.39 0.93 1.10 0.67 0.89 1.69 0.38 WO 03/066678 PCT/AU03/00137 49 Table 2.

Cluster 1 2 3 4 5 6 7 8 1 0.32 0.70 0.87 1.09 0.71 0.84 1.39 0.80 2 0.69 0.29 0.70 0.59 0.64 1.09 0.94 0.71 3 0.89 0.74 0.39 0.81 0.84 1.06 1.06 0.75 4 1.11 0.62 0.80 0.36 0.80 1.47 0.66 0.87 0.71 0.66 0.82 0.80 0.34 1.16 0.89 0.72 6 0.84 1.09 1.04 1.46 1.15 0.31 1.69 0.94 7 1.40 0.96 1.05 0.67 0.90 1.70 0.38 1.09 8 0.84 0.77 0.77 0.89 0.76 0.97 1.10 0.43 WO 03/066678 PCT/AU03/00137 Table 3.

Cluster 1 2 3 4 5 6 7 8 9 1 0.32 0.70 0.86 1.09 0.71 0.84 1.39 0.79 2.14 2 0.69 0.29 0.70 0.59 0.64 1.09 0.93 0.71 1.67 3 0.88 0.74 0.39 0.81 0.84 1.06 1.06 0.76 1.61 4 1.10 0.62 0.80 0.36 0.81 1.47 0.67 0.87 1.24 0.71 0.66 0.82 0.80 0.34 1.16 0.88 0.72 1.76 6 0.84 1.09 1.04 1.46 1:15 0.30 1.69 0.93 2.37 7 1.40 0.96 1.05 0.67 0.89 1.70 0.38 1.09 1.09 8 0.83 0.77 0.77 0.90 0.76 0.96 1.10 0.43 1.77 9 2.15 1.69 1.61 1.24 1.77 2.38 1.10 1.77 0.43 WO 03/066678 WO 03/66678PCT/AU03/00137 51 Table 4.

0.31 0.4 2 RMSD' 0.31 0.52 RMSD 3 0.31 0.6 2 RMSD' 0.3' 0.7' RMSD' 4 1 0.11 4 1 0.15 2 1 0.39 9 1 0.51 2 2 0.42 2 2 0.34 4 2 0.26 4 2 0.22 3 0.15 5 3 0.17 16 3 0.21 16 3 0.21 3 4 0.30 12 4 0.64 12 4 0.64 Aver 0.31 6 5 0.15 4 5 0.27 16 5 0.00 12 6 0.64 3 6 0.30 37 6 0.57 11 7 0.29 13 7 0.21 Aver 0.34 4 8 0.12 14 8 0.22 9 0.00 10 9 0.14 14 10 0.26 1 10 0.12 13 11 0.36 31 11 0.21 17 12 0.21 37 12 0.60 27 13 0.06 37 13 0.20 31 14 0.21 37 14 0.72 29 15 0.15 .35 15 0.88 37 16 0.42 Aver 0.34 37 17 0.45 19 18 0.71 1919 0.80 2 20 0.80 16 21 0. 78 Aver 0.35 Table 1 2 3 4 65 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 0 1 0.39 0.75 0.81 0.79 1.09 1.25 1.36 0.88 0.85 1.06 1.25 1.40 1.85 2,08 2.67 2.80 3.07 3.19 3.10 2.69 3.37 3.02 2.93 3.22 3.05 3.07 3.28 1.99 1.66 1.66 2.20 1.59 1.66 2.51 2.85 2.50 2.83 3.24 3.18 2 0.4 0.80 0.2 0.95 090 1.18 0.84 1.02 1.00 0.93 0.97 1.45 1.85 2.31 2.48 2.73 2.91 2.77 2.34 3.08 2.79 2.71 2.97 3.09 3.01 3.14 1.84 1.65 1.39 2.14 1.46 1.74 2.57 2.65 2.43 2.78 3.10 3.12 S 0.42 0.75 0.88 1.02 0.83 1.26 .32 1.23 1.36 1.18 1.61 1.48 2.21 2.32 2.73 2.79 Z.70 2.27 3.07 2.58 2.59 2.87 3.223.03 3.12 2.23 1.95 1.76 2.54 1.92 20 2.8B 2.88 2,78 3.09 3.20 3.38 C, 4 0.41 0.72 0.89 0.95 1.18 1.14 0.80 1.32 1.28 158 176 2.39 2.48 2.85 2.94 2.80 2.40 3.16 2.76 2.68 2.98 3.14 3.00 3.18 1.97 1.62 1.54 2.39 1.77 1.89 2.70 2.66 2.53 2.88 3.23 3.21 S0.44 0.74 0.87 1.22 1.28 0.93 1.14 1.03 1.17 1.54 2.02 2.11 2.45 2.58 2.39 1.99 2.74 2.41 2.34 2.62 2.99 2.77 2.87 1.81 1.60 1.30 2.24 1.66 1.892.66 2.59 2.39 2.74 2.88 3.02 0c 0.47 1.01 1.22 1.33 0.88 1.07 0.96 09 1 63 1.93 2.09 2.36 2.53 2.27 1.83 2.61 2.39 2.27 2.57 2.92 2.70 2.76 1.67 1.52 1.11 2.20 1.59 1.86 2.58 2.49 228 2.73 2.84 2.93 7 052 1.70 176 1.46 1.57 1.16 1.44 1.05 1.81 1.89 2.38 2.40 2.33 1.95 2.78 2,21 2.28 2.49 3.01 2.85 2.83 2.38 2.15 1.84 2.77 2.20 2.43 3.14 2.99 2.93 3.10 2.91 3.27 8 037 0.76 0.94 0.90 1.34 1.62 2.33 2.63 2.74 2.88 3.06 2.94 2.55 3.06 2.98 2.78 3.01 2,64 2.75 2.95 1.45 1.32 1.22 1.63 1.03 1.25 2.07 2.26 1.96 2.31 2.79 2.68 9 0.49 0.84 1.05 1.48 1.72 243 2.66 2.80 2.86 2.992.94 2.57 3.02 2.92 2.72 2.92 2.47 2.57 2.81 AO1.40 1.08 1.26 1.57 1.03 1.01 1.87 2.03 1.86 2.20 2.69 2.55 0.50 1.06 1.34 1.34 2.16 2.39 2.50 271 2.83 2.63 2.22 2.83 2.69 2.45 2.78 2.58 2.51 2.75 1.27 0.98 0.96 1.77 1.22 1.30 2.06 2.02 1.83 2.38 2.83 2.67 11 0.45 0.82 1.11 2.08 2.11 2.32 234 2.64 2.4 2.11 2.65 2.67 2.54 2.73 2,752.77 2.87 1.33 1.39 0.90 1.55 0.01 1.39 2.18 2.32 1.92 2.16 2.55 2.59 12 0.50 0.95 1.52 1.65 1.90 203 2.28 214 1.79 2.48 2.292.29 2.41 2.86 2.77 2.69 1.84 1.85 1.28 2.11 1.52 1.95 2.67 2.74 2.422.51 2.52 2.87 132 0.53 1.69 1.45 1.68 1.74 2.03 1.68 1.30 1.97 2.02 1.88 2.09 2.56 2.35 2.28 1.55 1.66 0.99 2.01 1.56 1.93 2.51 2.41 2.11 2.32 2.23 2.49 14 056 1.32 1.31 1.90 1.78 1. 1.82 2.35 1.54 1.80 1.89 2.43 2.36 2.29 2.64 2.70 2.333.00 2.67 2.93 2.96 2.86 3.182.84 2.42 2.84 0.43 0.81 0.79 0.88 (.89 0,86 1.34 1.10 1.20 1.17 1.67 1.67 1.42 257 2.61 2.16 2.42 2.50 2.71 2.24 2.22 2.50 1.93 1.41 1.90 16 048 1.03 0.92 1.02 1.05 1.41 0.88 0.95 0.83 1.43 1.39 1.35 2.65 2.61 2.32 2.49 2.66 2.75 2.08 2.03 2.45 2.00 1.50 1.90 17 0.43 0.71 0.73 1.05 0.92 1.19 1.23 0.97 1.26 1.39 1.07 2.41 2.62 2.30 1.95 2.40 2.41 1.76 1.85 1.99 1.35 0.82 1.35 18 0.44 0.80 1.17 1.09 0.87 1.08 0.77 1.03 1.20 0.87 2.53 2.58 2.60 2.022.61 2.41 1.61 1.66 1.96 1.45 0.92 1.38 19 044 076 0.78 1.04 0.95 0.95 1.33 1.22 0.90 2.50 2.58 2.29 2.25 2.65 2.64 1.89 1.78 2.08 1.62 1.07 1.36 0.49 1.09 1.17 0.94 1.28 1.74 1.46 1.29 2.30 2.33 1.94 2.54 2.44 2.62 2.30 2.01 2.37 2.07 1.55 1.77 0.49 1.32 1.09 1.10 1.21 1.04 0.81 2.24 2.41 2A40 1.98 2.55 2.38 1.63 1.45 1.66 1.34 0.92 0.93 21 0.41 0.84 0.86 1.23 1.10 1.052.662.52 2.61 2.37 2.83 2.58 1.81 1.652.151.89 1.44 1.66 U1 22 051 0.86 1.250.86 0.98 2.39 2.24 2.33 2.45 2.67 2.46 1.81 1.46 2.03 1.94 1.48 1.53 IQ 234 0.46 0.83 0.88 0.93 2.44 2.39 2.59 2.082.64 2.30 1.48 1.44 1.90 1.05 1.11 1.39 24 0,48 0.81 0.85 1.92 1.91 2.45 1.55 2.13 1.72 0.87 0.98 1.37 1.13 0.95 1.05 0,47,0.7B 1.99 1.91 2.40 1.90 2.36 1.96 1.21 0.88 1.48 1.49 1.25 1.08 26 2.11 2.17 2.5 1.83 2.442.13 1.30 1.14 1.50 1.27 0.89 0.90 27 GAS, 0.80 0.8D 0.95 0.87 0.99 1.36 1.37 0.83 1.44 1.97 1.60 289 .0.50 0.99 1.26 1.05 0.86 1.37 1.29 1.13 1.74 2.22 1.85 29 0.4 '1.38 0.93 1.27 1.90 1.88 1.44 1.93 2.39 2.18 31 4 0.46:0.67 0.89 1.01 1.39 0.69 0.89 1.43 1.36 31 0.46:'0.83 1.50 1.76 1.30 1.65 2.09 1.98 32 0.5.08 1.36 1.16 1.431.93 1.78 33 0.52 0.85 0.97 0.95 1.22 1.11 34 .5t 0.94 1.28 1.42 1.02 0:47: 0.99 1.42 0.98 36 70. 0.780.86 37 0.49 0.86 3 0.52 39.

39 0

C

C

C

-1 WO 03/066678 PCT/AU03/00137 Table 6.

Method Family TOL(A) TOL Sealevel peak) c, Cl Unique Avg. Intracluster RSD Avg. Intercuster I1SD O 0.25 0.25 0 744 690 0.16 2.10 o 0.5 0.5 0 4862 4862 0.21 2.26 O 0.75 0.75 0 8002 889 0.27 2.37 C 0.25 0.2 0.125 2427 1748 0.16 2.10 G 0.25 0.2 0.25 1005 1407 0.16 2.10 G 0.25 0.2 0.5 812 626 0.16 2.10 G 0.2 0.2 0.75 223 166 0.23 2.11 G 0.25 0.3 0.125 3801 2687 0.18 2.11 G 0.25 0.3 0.25 2498 1825 0.17 2.11 G 0.25 0.3 0.5 954 709 0.16 2.10 G 0.25 0.3 0.75 281 217 0.22 2.10 0.25 0.5 0.125 3802 2688 0.18 2.11 G 0.25 0.5 0.25 2498 1825 0.17 2.11 G 0.25 0.5 0.5 956 711 0.16 2.10 G 0.25 0.5 0.75 295 225 0.21 2.10 G 0.5 0.2 0.125 2079 2079 0.19 2.26 G 0.5 0.2 0.25 2073 2073 0.19 2.26 G 0.5 0.2 0.5 1853 1853 0.18 2.26 G 0.5 0.2 0.75 1047 1047 0.17 2.26 G 0.5 0.3 0.125 7654 7654 0.23 2.26 G 0.5 0.3 0.25 6715 6715 0.22 2.26 G 0.5 0.3 0.5 4071 407 0.19 2.26 G 0.5 0.3 0.7b 1540 1540 0.16 2.26 G 0.5 0.5 0.125 10135 8991 0.29 2.26 G 0 0.5 .5 0.2 7413 7013 0.26 2.26 C 0. 0.5 0. 5 1 4088 4088 U.19l 2.26 O 0.5 0.5 0.7 1540 1540 0.16 2.26 G 0.75 0.2 0.125 1848 1848 0.21 2.37 G 1.75 0.2 0.25 1848 18418 0.21 2.37 C 0.7E 0.2 0.5 1846 1846 0.21 2.37 C 0.7,1 0.2 0.75 1646 1040 0.21 2.37 0 0.71 03 0.125 8040 804) 0.25 2.37 C 0.7 0.3 0.25 7973 7973 0.25 2.37 C 0.71 1.3 0.5 710 714 0.14 2.37 C 0 751 0.3 0.7 1215 1 121 (1.21 2.37 0.71 0 .12 33317 1401 F 0.67 2.32 0 0.75 U.5 1)25 J 113 11479 0.51 2.3-1 C 0.75 0.5 0.5 9975 8506 0.31 2.37 c 0.75 .5 0.75 1726 -1598 0.131 2.38 Table 7.

SI I 4i 0 j I 9 Q o 9.9 (210 911 926 Q11 91I 12 16 I91 7 I Q2,9 I9 Q234 I QlQ8 92 Q6 Q2 1 I28 IQ19 I 9306 Sim If Cof 6816 8271 28 609 1809 90 31 274 2 1 *12 39:3 117,5 13 10 30 1667 11648 1 3861 882 1 428 1855 1473 1545 18611 (21 Q2 I Q3 Q] Q5 0 7 1 Q5 91 33213 2.56 3.72 3.09 0.69 2.32 3.14 3.M 9 Q& 0.13 3. IX? 1 1.7 2.08 2.04 1.79 2.23 Q93 -177 1.91 0.27 1.9 3.77 '1983 1.7821.

1 3 O 1.71 1.51 1.43 3.16 I.N 3.0(C 1.47 170 2.56 3.71 3.96 0.71 2.23 3.11 :.91 Q 121 2.45 13.97 3.24 2.3 1 1.13 3.23 14.1 (J9 gio 1)(2.11 1i12 Q13 1Q -0,16 1 1 QO1o 16 1 Q22 194j1 jq4 Q21[I2Q261]Q3 I 18 T 7.481 293 990 1' 07 312 2.62 1.9 2.41 2.31 2.3112.84 2.31 3.86 2.37 2.1212.96 2.40 2 01t9 1. 2.83 I 7 .3 2.54 [21 .13 2. 3 2.98 1.11 1.45 1.47 1.07 2.54 2.54 1.75 2.73 2.21 1.73 1.61 101 2.47f 1.22 2 T36 29 3 .71 3.26 4.16 1.85 2,72 2.68 1.66 3-98 3.9912.59 2.67 1.08 2.67 2. 2 1.201.86 -12 1.791.56 1.84 2.11 3.16 2.33 321 119 1 1I7 1.73 1.55 3.16 3.16 1.85 2.20 1.40 2.21 2 1.55 1.60 I15 2.72 3.90 2.51 2090 2.33 3J5 2.80 2.14 1,77 1.95 2-39 2.32 2.32 2.81 2.33 3.84 2.33 2.0812.96 .31 2.81 '3.34 2.74 2.28 1I. 2.19) 2. 5 12 2. 76 212 2.14. 2.74 1.1 1.11 2.44 2.28 4.06 2.28 1.97 2.90 2.73 3.30 1.75 1.47 1.72 2.29 3.12 2.19 2.76 3.34 1.41 1.75 2.01 1.44 3.12 3.12 1.99 2.15 1.45 2.15 2.3011.61 1.44 Q7 I 3.13811.09 9813137R1.20 1.08 1.0413.3 3.112 27 1.17 1.0 16 39 :1 IA9 tj2 2.33 1 (.90 1 2.36 3 2.98 1 -1.12 3.02 1 31 1 4.19 1 2.19 1 2.76 2.84 12.25 i 4.12 I4.12 1 2.39 I 2.73 I 1.28 12.73 12.89 1.79 12,25 S12q 0 2.81 4.00 3.43 1.30 2.37 3.30 4.20 i1112 2.77 1 289 303 23 L413 2.45 '277 2.18. 1.90 2.73 2.36 2.36 2.76 2.53 4.14 2.32 3.27 2.73 QIO 2.69 L88 12.45 1.77 2.6P3 1.64 1.7:3 2.26 2.77 0120 2.28 1.18 2.12 261 J.80 2.201 2.66 J.98 1.72 1.59 1.98 2,64 2.65 1.06 2.60 2.09 1.59 1.47 2.13 1.9 911 :199 2.26 1.23 1.56 3.99 4.11) 1.49 0.96 120 2.33lO23 2.34 3.05 4.11 2. 910 331 4.32 2.16 2.85 2.82 2.23 4.11 4.11 2.46 2.60 121 2226 3.07 1.76- 2.23 91' -29 A 1.75 2.33 1.74 2.45 2.61 1.i5 .29 2.80 L.T 227 0.6 1.89 2.1 2.11 2.1712.76 1.83 1.36 1.0 1.86 2.61 2.62 1.42 1.23 2.09 1.22 1.53 2.01 1.89 91.? .8 2.32 7 2.1(1 2.88 2.12 2,21 297 .104 2.11 -3(00 1.86 0.28 212 1.07 2.26 1.6 2.61 2.22 2.24 2.63 2.13 2.13 1.93 1.82 2.97 1.82 2.00 2.50 2.63 14 '.38 2.58 4.07 3.06 2.38 1.10 2.98 44 "143 '27I 411 '267 192 111 S6') 2.31 1.30 2.85 1.96 209 2.83 1419 1.0 2.55 2.22 4.00 2.23 207 3.14 2.8 915 3.04 2.34 2.71 2.33 3.04 2.11 2.10 3.01 2.89 1.89 2.95 2.02 10 2.11 0.28 2.23 1.89 2.63 2.28 2.22 2.61 2.11 2.11 1.74 2.01 2.99 2.01 1.89 2..10 2.61 Q16 3.26 9.5 25 2.80 3.26 2.52 2.76 3.5, 7 1 1 2.19 3 1 2.22 2.28 2.52 2934 0923 2.72 2.88 2.56 2.59 2,8 .51 2.51 2.39 2.36 2.96 2.36 2.28 2.7 2.86 Q1, 7 T 712.9 4.13 3.17 2.71 1.49 3.30 4 _1 241 2,64 4.23 2.79 1680 1.46 101 '2.75 0.64 3.19 2.40 2.17 3.17 1.48 1.49 245 2.40 4.18 2.40 2.20 3.51 3.17 1.8 2.50 1.14 1.8D 1.39 2.50 2.83 1.29 2.12 2.76 1.97 2.12 1.71 .71 2.83 2.70 2.87 3.30 0.37 1.43 .44 0.86 2.83 2.83 2.1) 2.01 1.91 2.01 1.87 1.19 0.86 Q19 18 1f.50 2310 1.96 1.83 2.12 1.74 2.75 2.19 1.71 2.22 1.43 2.22 2.12 25 2.43 1.47 0.30 1.04 IU .1 1 1.76 17 2.75 1.47 1.20 1.90 1.46 9/U 2.03 1.52 2.00 1.72 2.03 2.16 2.01 2.78 1.91 1.59 1.61 100G 2.22 218 2.21 '238 2.22 1.46 1.05 0.28 1.48 2.18 2.18 1.56 1.62 2.70 1.62 1.35 2.65 1.42 91 2.42 1.1-4 1.81 I.A 2.42 2.78 1.38 2.20 2.76 192 2I 1.85 2.61 2.8 2.72 6 3.36 0.69 1.37 1.58 0.00 2.78 2.78 2.16 2.01 1.89 2.01 1.99 1.25 0.56 219 2.47 3.98 3M17 2.19 1.0 3.01 4.23 2.34 278 4.20 21 2 _91 1316 2.21 239 1.71 '2.72 1.90 2.12 2.08 1.07 1.07 2.72 2.24 3.92 2.25 2.09 2.93 2.7 30 2. 2.47 13.!T 3.08 2.30 .1 3.05 4.24 2.27 2.69 4.24 2.71 2.25 1.12 2,25 2.30 1.67 2.75 2.10 2.08 2.71 1.11 1.11 2.66 2.34 3.93 2.34 1.97 2.89 2.71 9'4 2.83 1.76 2.Fi8 1.85 2.83. 2.522. 8 166 _41 1.51 1.91 2.3 1.77 23) 2.49 2.06 1.79 1.56 2.07 2.63 2.53 0.26 136 2.27 1.36 1.25 1.99 2.08 Q"8 2,11 1.71 2.65 2.23 2.41 2.20 2.21 2.02 2.57 1.53 2.62 1.31 1286 216 102 2-.5 2.35 2.06 1.47 1.68 2.02 2.27 2.27 1.20 0.71 2.72 0.68 1.17 1.94 2.02 a26 3-88 '.17r 1.0' 1.12 3.6) 3.94 1.49 1.26 16 211 1 9j 935 9 3.133 1 .20 1.93 2.50 2.73 1.95 3.94 3.94 2.29 2.69 0.22 2.69 2.98 1.21 1.95 927 2.37 1.67 2.62 2.28 2.37 2.23 2.2 2.52 26 '6 1 1.3 7 23 I 09 2001) 2.38 1.98 1.30 1462 1.98 2.24 2.24 1.27 0.68 2.79 0.70 1.12 1.89 1.98 7 '1-4 1.6-0 2.8 2.42 2.14 .09 2.30 14.80 2.13 1.47 3.02 1.61 00 2.28 2.25 1.9 1 1.21 1.35 1.98 2.0 2.09 1.24 1.25 2.96 1.27 0.28 1.91 1.98 31 1.1 1.2. 1..1 7 .53 1.32 1.7 2.08 1.34 3.02 3.02 .9 0.9 111.33 3 2.1 33.2 3.31 [531 242I .7 I3:7I22 2721.9I27 1.0 22I2732.9 .8 .341.88 I 1.4017 35 I 2. 77 II. .2.77219 I~s 2.03 1 1.88 I .03 I 2.01 11.25 311( 11 -2.42 1 J.12 1 1 1-50) 1 2.42 11 2.7V 1 1 3 1 2-21 2.72) 1 L95 1 2-'7J7 I I.S51 1 2-62 1 2.77 1 2.69D 2-811 3.34 1 0.69 1 1.40 1 1.57 1 0.53 1 2.77 1 2.77 1 2-19 1 2.03 1 1.88 1 2.031 201 11,2 I I I I I I I I I I I I I L_ WO 03/066678 PCT/AU03/00137 Table 8. he I ail IO [1a c ;ek G I~C~i Unique F" ICI fAvg. nimrItr stor RMSD I Avg. lnterduster RMSD 0) 0.253 0-25 0 6 0.31 2.46 0o 0.5 0.5 0 2U____55 3 0.23 2.58 o 0.7 5 01698 4698 0.29 2.59 G -0.25 0.2 0.12b 88 38 0.35 2.18 0.25 87 87 03 2.48 C 0.2 U.2 0.2 78 78 0.5 4 G 0.25 0.2 0.5 .57 G7 0.20 2.48 G 0.25 0.3 0.26 475 322 0.25 2.45 G 0.25 0.3 0.25 4,38 307 0.2 2.46 6 0.25 0.3 0.5 289 223 0.25 2.46 G .2 030.75 131 1.4 0 28 24 O 0.28 -05 0,125 498 342 0.23 2.45 G 0.25 050.25 453 329 0.23 2.45 G 0.25 0.5 0.5 309 237 0.24 2.4 G -0.23 0.5 0,75 162 135 0.25 2.46 G 0.5 0.2 0.125 85 63 0.23 2.59 G 0.2 0.25 G9 68 0.23 2.59 G 0.5 0.2 0.5 C8 08 0.23 2,59 -G 0.5 0.2 0.75 57 57 0.22 2.80 0.3 0.125 1890 270 0.22 2.8 GC 0.5 0.3 0.25 1848684 0.22 2.58 G050.8 0.5 5252 1 5[252-9 0.21 2.58 G0.85. .5 5 5 0.27 2.58 0.8 0.125 4258 4258 0.72.8 IC0. 5 0.5 0.'3016 3010 0.24 2.5 (5 0.5 0.5 1800 165r006 0.21 28 0.3 0.75 51251 525 0.2)2 2.58 O0.75 0.2 0.2 45 40 0.15 2.08 el 0.75 0.2 0.25 45 45 0.11 2.585 C0.75 0.2 0.4354 0.15 c; 0.75 0.3 (125 155.4 1 181 02 1 0,75 0.1 30.)5 185.3 11M C0. 7 0.3 18(10 1806 0.1 0.3 0.75 12. C .71 0.5 0.115 609 5 0.7 0.25 :3600 :3660 0.3 1G 0.7T7 0.1 0.11 10(1 1007 0.75 0-I 05 0.75 110 13490 2 1.0(.12-,11 70 0(.46 2.55 0.2 0.7 107 10 Table 9.

iL2L i 0 Q1 Q5 06 0ii2 L09 010 012 012 Q01 014 IS 0161 Q17 018 018 90 021 0V8 Q29 Q24 025 Q86 Q3' Q28 I29 Sizeof C4,stJT 298 238 821 309 2-15 174 117 168 47 2(1 587 3413 281 ]l 182 270 185 329 19 308 772 716 232 702 239 650 401 284 298 tl 2 Q Q QJ Q9 f;9 QJS I Q1 4 QI Qj;6 1171 Q Q19 QRO 1 (2-10 Q£1 022 1 0RY9 Q04 5 IW Q2 I f (0 fo 1 0.6.9 2.02 2.6 1.47 2.67 1.90 3.14 2.0 2.18 2.71 272 2.59 2.24 2.24 2.80 3.10 2.2 3.22 2.45 2.49 2.23 2.03 1.83 2.79 1.83 2.92 2.01 2.45 2.26 2.17 q2 2.00 0.C6 2.46 2.44 3.20 2.;31 4.07, 2.4 1.15i 2.48 .31 2.68 2.52 2.37 3.21 4.013 2.14 3.96 2.79 2.83 2.08 2.4 2.28 2.86 2.27 3.03 2.43 2.80 2.56 2.45 Q8 1.9-1 2.43 0.99 1.,5 2.41 2A3 3.03 2.23 2.25 2.65 2.16 2.74 2.35 2.15 2.03 2.99 2.18 2.85 1.94 2.44 2.36 0.98 1.87 2.38 1.30 2.53 0.95 1.95 2.63 1.90 (24 14. 2.42 1.83 l.73 2.70 2.17 3.03 1.9 2' 23.7 2.;5 2. 63 2.1 1.8 1.6 p. 1.9 97 3.115 2.48 2.67 2.20 1.83 2.02 2.87 2.02 2-90 1.80 2.48 2.26 2,9 2.7 3.37 2.44 2.80 0.57 3.00 4.2, 3.71 3.42 2.82 3.13 2.07 2.63 2.71 3.20 4A1 3.10 4.48 3.73 4.09 3.03 2.43 2.21 4.09 2.21 3.94 2.37 3.73 3.78 3.59 Q6 2.03 2.25 2.47 2.25 3.30 0.30 2.85 1.04 2.52 3.60 2.15 3.54 2.50 2.60 2.38 2.47 2,50 2.60 2.26 1.97 2.44 2.47 2.27 2.35 2.27 2.45 2.51 2.25 1.77 2.08 Q7 3.20 4.06 3.01 3.ii 4.32 2.86 0.26 2.84 .15 4.93 1.8 4.74 3.01 3.62 1.641 1.3 3.96 1.29 2.24 1.89 3.95 3.01 2.99 1.92 2.99 1.90 3.02 2.24 2.45 2.45 2QS 2.15 2.47 2.28 2.07 3.70 1.08 2.83 0.30 2.36 3.64 2.36 3.,4 2.02 2. 1, 2.31 2.64 2.46 2.44 2.03 2.11 2.83 2.29 2.45 2.44 3.44 2.38 2.33 2.03 1.76 2.29 Q9 2.27 1.27 2.31 2.39 .36 2.53 4.16 2.30 0.34 244 3.17 2.57 2.36 247 3.82 4.05 2.15 4.09 2.71 3.15 2.22 2.31 2.48 3.09 2.48 2.90 2.29 2.71 2.69 2.67 1QO 2.056 2.40 2.4 7 .5 2.80 2 4. 3 3. 2 2(1 0.71, 3.95 114 2.01 244 (.91 I.0 2.28 4.91 3.51 3.99 2.06 2.66 2.73 4.07 2.73 4.14 2.61 3.52 3.49 3.39 011 2.78 3.29 2.25 2.C; 3.1.7 2.15 1.84 2.35 3.17 .95 0.30 3.983 2.81 2.93 1.07 1.79 3.24 2.01 2.18 2.32 3.13 2.25 2.06 1.71 2.06 1.35 2.26 2.1.8 2.33 2.32 WI 9 2.60 2.9 2.65 2.69 3.74 4.36 3.73 2.45 1.26 3.97 1.10 2.47 242 3.98 4.01 2.04 4.85 3.60 3.87 2.21 2.68 2.08 4.15 2.68 4.09 2.64 3.60 3.38 3.27 (11 2.10 239 2.31 2.22 2.2 2.07 3.50 2.69 .29 5 8 1 2 .28 0.64 1.20 2.90 3.74 1.&2 .3.05 2.07 2.73 1.03 2.31 2.12 2,83 2.12 2.72 2.27 2.67 1.96 2.62 i .1 2.1 2.32 2.12 2.04 2.71 2.60 3.57 2.07 2.42 2.45 2.91 2.3C 1.21 0.61 2.82 3.60 1.59 3.74 2.67 2.90 1.79 2.12 2.24 2.70 2.24 2.86 2.07 2.07 1.94 2.50 2.78 3.16 2.06 2.50 3.29 2.38 1.83 2.12 31.32 3.93 1.08 3.96 2.9C 2.80 (.28 2.04 3.11 1.77 2.37 2.17 3.21 2.06 2.25 1.35 2.25 1.62 2.06 2.37 2.33 2.20 02 3.25 4.02 3.00 3.28 4144 248 1.32 2.64 140 4.91 1.79 4.83 3.79 370 2.0( 0.27 3.98 1.01 2.14 2.01 4.04 3.07 2.03 1.89 2.92 2.02 3.09 2.13 2.05 2.07 Q17 2.13 2.18 2.24 1.91 3.03 2.54 3.94 2.43 2.10 2.28' 3.24 2.25 1.77' 1.53 3.10 3.97 0.53 4.0 2.69 2.68 1.18 2.23 2.37 .10 2.37 3.22 2.21 2.69 2.23 2.37 Q18 3.28 3.94 2.91 3.23 4.50 2.66 1.29 2.44 4.07 4.91 2.02 4.82 3.71 3.,9 L75 1.07 4.01 0.26 2.06 1.88 3.97 2.91 3.02 1.99 3.02 1.92 2.93 2.05 2.53 2.32 Q19 2.49 2.80 2.05 2.52 3.87 2.25 225 2.06 2.65 3.2 .29 3.683 2.70 2.69 2.45 2.24 2.66 2.15 0.70 1.28 2.49 2.05 2.24 2.04 2.24 1.89 2.09 0.70 2.04 1.18 (Q9 2.5 2.88 2.6 2.72 4.12 1.97 1.89 2.12 3.15 3.99 2.31 3.96 2.7 2.94 2.18 2.00 2.72 1.87 1.41 0.28 2.89 2.66 2.47 2.00 2.7 2.00 2.60 1.40 1.88 1.33 Q1 2.17 2.09 240 2.1-1 2.95 240 3.93 2.49 2.21 2.08 3.11 2.33 1.57 1.75 3.18 4.02 1.17 3.95 2.51 2.85 0.57 2.39 2.26 3.26 2.26 3.11 2.37 2.52 2.22 2.88 Q2 1.98 2.61 41. 1.81 2.32 2.43 2.91 2.28 237 2.61 2.21 2.73 2.36 .10 2.12 2.13 2.38 2.75 2.01 2.58 2.52 0.96 1.35 2.57 1.35 2.59 0.111 2.02 2.47 1.99 Q21 1.8.4 021-2.74 2.42 1.36 2.01 2.10 2.30 2.00 2.46 2.92 2.52 I 2.85 4.09 2.36 1.91 2.45 2.75 2.02 2.63 2.16 4.12 I 1.71 4.01 2.95 2.26 2.2 2.79 2.50 2.891 2.18 2.44 2.43 1.36 I 0.93 2.61 0.92 2.50 1.33 2.19 2A.46 1.79 2.70 1.89 0.14 1.98 2.0612.0713.26 2.52 2.1 0.27 2.5011.0812.54 2.06 2.20 1.85 095 1.50 2.34 1.38 2.100 2.10 2.16 2.87 2.33 2.87 3.07 2.50 2.88 3.95 245 1.91 2.39 Q27 2.03 2.5,4 0.94 1.83 2.28 245 2.90 2.2.4 028 2.49 2.80 2.05 2.52 13.S7 2.25 2.25 2.00 (Q29 2.35, 2.43 2.53 2.37 3.83 1.81 2.44 4.81 4QSO 2.2.1 2.53 2.3 1 2.66 13.212.09 2.A3 2.32 .45 2.74 1.94 2.04 2.14 2.24 2.-1 2.79 2.39 2.0 2 2.2.3 2.33 1.38 0.94 2.05 0.95 2.01 1.32 2.23 2.50 1.80 91 4.10 1.36 4.09 2.77 2.90 1.02 1.99 3.21 1.L194 1.96 2.02 3.15 2.50 2.3 1.08 2.53 0.28 2.54 1.97 2.20 2.04 2. 56 2.2-1 2.69 2.32 2.11 2.04 2.90 2.30 2.82 2.04 2622. 44 0.93 1.36 2.61 1.35 2.68 0.87 2.04 2.51 1.98 15 3.52 2.29 3.04 2.70 2.9 2.45 2.24 2.67 2.1 0.70 1.27 2.49 2.05 2.24 2.04 2.24 1.80 2.09 0.70 2.04 1.18 70 3.47 2.33 3.25 2.04 2.01. 2.32 2.34 2.29 2.52 2.06 1.88 12.29 2.53 .53 2.20 2.53 2.21 2,54 2.00 0.30 2.14 37 3.411 2.32 3.39 2.60 2.18 2.20 2.07 2.42 2.32 1.81 1.32 2.63 2.11 1.95 1.86 1.95 2.05 2.11 1.31 2.14 0.31 Table j Method Family ?OL TOL (A) Sealevel peak) yEi 1 ciC Unique 7-o Ii1 I Avg. Intracluster RMSD I Avg. Intercluster RMSD 0o 0.5 0.5 0 886 886 0.28 2.83 0 0.75 0.75 0 2364 2364 0.32 2.62 G 0. 5 0.3 0.125 203 203 0.30 2.84" G 0.5 0.3 0.25 201 201 0.30 2.84 G 0.5 0.3 0.5 188 188 0.31 2.84 G 0.5 0.3 0.75 108 108 0.37 2.85 0 0.5 0.5 0.125 1913 1913 0.31 2.83 G 0.5 0.5 0.25 1197 1197 0.28 2.83 0.5 0.5 568 568 0.27 2.84 G 0.5 0.5 0.75 194 194 0.37 2.84 G 0.5 0.7 0.125 2038 1975 0.32 .2.83 G 0.5 0.7 0.25 1199 1199 0.28 2.83 C 0.5 0.7 0.5 568 568 0.27 2.84 G 07 G 0.5 0.7 0.75 194 194 0.37 2.84 G 0.75 0.3 0.125 175 175 0.33 2.65 G 0. 75 0.3 0.25 175 175 0.33 2.65 G 0.75 0.3 0.5 r 175 175 0.33 2.65 G 0. 75 0.3 0.75 162 162 0.34 2.65 G 0.75 0.5 0.125 2759 2759 0.32 2.62 G 0.75 0.5 0.25 2516 2516 0.31 2.62 G 0.75 0.5 0.5 1601 1601 0.28 2.62 G 0. 75 0.5 0.75 728 728 0.26 2.62 G 0.75 0.7 0.125 4067 3811... 0.42 2.61 G 0.75 0.7 0.25 2916 2916 0.35 2.62 G 0.75 0.7 0.5 1621 1621 0.28 2.62 G 0.75 0.7 0.75 729 729 0.26 2.62 Table 11.

Q1 Q Q' Q4 5 QG q7 Qs Q QI Q11 Q15 QI3 QQ5 Q.1 6 QIG Qi- Q1 9 QiP QO QS1 Q22 Q23 Q4 Q QSG Q27 Q8 Q29 Si I cf IC l0str 101 11:0 1 I112 11 91 115 19 i 5 200 184 1.54 13 I 1 97 11 188 270 104 11i 125 ,122 1 13W 1 2356 198 156 140 143 Q1 2 c Q 4 Q J5 Q f; gr P 1 w8 Q, Q10 Q11 I 12 I gl? I 14 I QJ.5 I| 016 (11- Q18 Q 19 (120 Q21 Q QS Q24 I 23 IQ29 Q27 IQ28 I4 1L 0.31 3.46 3.16 2.24 2.40 1.94 2.55 2.91 2.13 2.05 3.80 2.680 2.73 2A3 2.52 .76 2.32 2.33 2.13 2.63 2.43 2.40 3.14 1.33 3.12 2.13 2.41 2.21 2.55 2.53 Q2 0.37 2.11 2.88 2.27 27 2. 2.73 j3'G 3.85 1.94 2.50 2.419 2.07 2.12 .12 52 3.01 2.69 4.12 252 2.6 2.21 3.25 2.19 2.05 2.34 233 3.10 3.15 3.17 2.11 0.8 3.38 2.1 3.-1I1 3.02 2.73 l 4.04 2.33 2.33 2.32 '2.97 3.07 4.75 2.85 3.52 2.72 4.74 2.84 3.33 2.02 2.9,1 2.06 2.71 2.78 2.70 3.85 3-86 (4 2.23 2.88 3,33 0.3I 2.33 1.05 1.85 2.53 2.0 2.37 3.05 2.80 2.7. 2,15 1.8 2.99 1,32 1.93 2.31 2.95 1.70 1.10 2.78 2.76 2.92 2.30 2.27 2,33 2.43 2.20 2.40 .26, 2.60 2.33 0.37 2.09 2.53 2.5Z 3.12 2.90 2.74 2.08 2.25 2:37 2.59 3.57 ',15 2.23 2.34 3.47 2.38 2.47 2.49 2.35 2.37 2.35 1.48 1.63 2.99 2.83 (Q6 1 93 2.98 3-38 1.65 2.08 0.34 2.53 2.G 2 2.17 2.66 3.2 2.80 2,741 2.61 2. 9 2.65 L51 2.05 2.26 2.59 1.86 1.96 3.01 2.43 2.98 2.26 2.00 2.18 2.17 1.89 (Q7 2.55. 2.96 2.98 1.84 252 2.5.1 0.39 1.35 3.51 3.33 2.84 2.18 2.19 1.12 2.52 3.89 1.85 2.52 2.67 3.85 1.06 2.07 2.24 3.00 2.22 2.66 2.64 2.46 3.38 3.22 QS 2.93 2.73 2.72 9.55 2.55 2.67 1.35 D.39 3.83 3.89 2.65 2.14 2.17 1.3q9 2.70 .32 1.82 2.72 2.70 4.30 1.83 2.55 2.18 3.04 2.13 2.069 2.94 2.85 3.49 3.46 Q9 2.35 3.04 4.12 2.75 3.11 2.68 3.52 3.83 0.27 143 4.62 3.49 3.39 3.59 2.30 1.65 3,04 2.43 2.07 1.35 35.15 2.94 3.57 1.95 3,68 2.07 2.80 2.87 '2.11 2.16 QI 206 3.85 4.04 238 2. '91 2.0 3.34,1 3 185 1130 t 3.59 3.49 3.48 2,40 1.87 2.97 2.42 2.38 2.04 3.07 2.60 3.82 2.05 3.93 2.35 2.52 276 2.03 2.14 QJJ 3.75 1.85 2.24 3.00 2.70 3.16 2.85 2.1t 4.58 4.51 0.75', 2.48 2.61 2.69 3.'24 1.99 2.49 3.34 3.31 4.91 2.33 2.80 2.36 3.85 2.19 3.3U 2.91 2.73 410 4.14 Q12 28 5 2.43 2.27 2.82 2.01 2.74 2.14 2.093 3.50 3.6 2.51 o0.67 1.20 2.32 2".5i 4.07 2.36 2.67 2.63 3.95 2.15 2.64 1.72 2.66 1.49 2.63 1.98 2.06 3.21 307 Q11 280 2.44 2.25 2.7 2. 22 2.70 2.20 2.10 3.7 3.45 2.608 1.22 0.61 2.13 2.80 3.99 2.21 2.47 2.61 4.10 2.23 2.76 1.51 2.66 1.608 2.60 2.26 1.86 3.08 3.24 Q14 2.41 2.96 2.97 2.14 2.36 2.62 1.11 1.37 3.58 3.8 2.75 2.31 2.15 .40 2.58 3.90 2.01 2.34 2.66 3.93 1.87 1.84 2.21 3.03 2.40 2.65 2.52 2.52 3.26 3.42 Q. 2.5 1 2.44 3.07 1.88 2.60 2.25 2 2.52 2.70 9.3 2.40 3.30 2.81 2.861 2.59 0.34 2.73 1.80 2.18 1.65 2.72 .78 1.93 2.83 2.20 2.80 1.64 2.20 2.27 2.00 1.94 Q26 2.77 4.11 4.73 3.00 3.60 2.66 3.89 4.30 116 .88 5.01 .10 4.01 3,01 2.71 0.27 .31 2.906 2.61 1.08 3.43 3.00 4.23 2.52 4.25 2.62 3.00 2,85 2.04 1.79 Q17 2.31 2.52 2.84 1.32 2.16 11 1 1.8 1.82 3.04 3.00 2.57 2.40 2.24 2.00 1.80 3.41 0.36 1.92 2.01 3.40 1.09 1.58 2.26 2.58 2.31 2.01 2.28 2.30 2.83 2.66 Q18 2.29 2.98 3.48 1.93 2.16 2.00 2.51 2.67 2.42 2.45 3.36 2.68 2.47 2.31 2.18 2,94 1.84 0.65 2.19 2.86 2.08 2.07 2.80 2.44 2.86 2.19 2.37 2.44 1.79 2.09 Q19 2 2 .61 2.66 2.26 2.32 2.20 2.62 2.04 2.18 2.54 3.36 2.60 2.58 2.77 1 53 5 1.94 2.17 0.74 2.71 2.01 2.40 2.50 1.81 2.38 0.72 1.96 2.14 2.20 2.20 2.64 4.11 4.72 2.96 348 2.60 3.86 4.29 1.3 2.08 4.9 399 412 3.93 2.70 1.08 3.41 2.90 2.71 0.27 3.41 3.03 4.25 2.53 4.21 42.721 2.86 300 1.77 2.02 Q21 2.42 2.52 2.83 1.69 2.39 1.84 1.98 1.84 3.11 3t07 2-42 2.20 2.28 1.R 1.78 3.42 1.08 2,13 2.07 3.41 0.39 1.29 2.29 2,57 2.21 2.00 2.43 2.23 2.68 2.83 Q'22 -40 2.86 3.32 1.10 2.48 L,96 2.07 2,51 2,94 2.60 2.86 2.05 2.71 1.85 1.91 2.99 1,59 2.11 2.37 3.03 1.31 0.33 -2.92 2.78 2.76 2.36 2.34 2.24 2.24 2.45 Q( .3 16 2.19 2,02 277 2,49 3.0)2 2.25 2.14 3.55 3.83 2.41 178 1.55 2.20 2.81 4 2 2,26 2.79 2.53 4.24 2.28 2.92 0.4L 3.02 1.11 2.52 2.48 2.25 3.40 3.47 Q24 1.34 3.26 2.96 2.76 235 2,47 3.00 3.06 198 207 3.91 2.71 270 3.03 2.20 2.51 2.59 2.51 1.89 2.52 2.59 2.77 3.03 0.32 3.02 1.89 2.29 2.24 2.26 2.24 3.12 2.17 2.00 2.92 2.37 3,02 2.22 2.138 3.67 3.93 2.26 1.68 1.73 2.41 2,79 4.24 2.31 2.86 2.46 4.20 2.20 2.75 1.11 3.01 0.38 2.45 2.32 2.43 3.47 3.41) Q26 2.11 2.61 2.65 2.27 2.32 2.21 2.63 2.641 2.18 2.55 3.37 2.00 2.55 2.78 1.53 2.57 195 2.18 0.72 2,73 2.02 2.41 2.48 1.80 2.38 0.72 1.97 2.16 2.21 2.21 (27 .3 2.31 271 2.33 144 1,0 2.60 2.00 2.85 2A7 2.98 1.99 226 218 2,14 2.99 2.2 2.33 1.91 2.81 2.44 2.30 2.47 2,22 2.30 1.91 0.56 1.12 2.52 2.30 Q8 122 2,34 269 2.35 1.04 2.18 2.47 284 2.87 2.S0 2.14 102 2.54 2.5 2.8 2.29 2.49 2.07 3.00 2.23 2.25 2.26 2.24 2.47 2.07 1.21 0.34 2.28 2.47 _U29 2.56 3.1 3.86 2,4 3.01 2.18 3.37 3-90- 2.12 2.03 4.08. 327 3.11 3.27 1.95 2.02 2.82 1.87 2.28 1.77 2.69 2.24 3.42 2.25 3.48 2.23 2.50 2.29 0.29 1.09 2.53 3.1 3.8 21 284 1.(1 3.22 3.1 10 2.13 PI2 3.13 3.29 3.41 1 .4 1.79 2.16 2.29 2.02 2.82 2.46 3.48 2.23 3.41 2.29 2.27 2.47 1.08 0.31 Table 12.

Method Family TOL TO L Sealevel peak) C IJ Unique E I Ci Avg, Intracluster RMSD Avg. IntercluterRMSD 0 0.5 0.5 0 414 407 0.32 2.73 O07 0.75 0 1227 1227 0.34 2.60 G 0.5 0.3 0.125 75 75 0.22 2.74 G 0.5 0.3 0.25 75 75 0.22 2.74 G 0.5 0.3 0.5 73 73 0.22 2.74 C 0.5 0.3 0.75 57 57 0.18 2.74 G 0.5 0.5 0.125 799 774 0.31 2.73 G 0.5 0.5 0.25 626 607 0.30 2.73 CG 0.5 0.5 0.5 336 328 0.31 2.74 G 0.5 0.5 0.75 132 131 0.35 2.73 C 0.5 0.7 0.125 1008 977 0.33 2.73 G 0.5 0.7 0.25 713 690 0.31 2.73 G 0.5 0.7 0.5 343 335 0.31 2.74 CG 0.5 0.7 0.75 133 131 0.36 2.74 G 0().75 0.3 0.125 75 75 0.37 2.63 G 0.75 0.3 0.25 75 75 0.37 2.63 G 0.75 0.3 0.5 75 75 0.37 2.63 G 0.75 0.3 0.75 72 72 0.31 2.63 G 0.75 0.5 0.125 924 924 0.32 2.61 G 0.75 0.5 0.25 875 875 0.32 2.61 G 0.75 0.5 0.5 632 632 0.30 2.61 G 0.75 0.5 0.75 304 304 0.34 2.61 CG 0.75 0.7 0.125 1874 1874 0.37 2.60 G 0.75 0.7 0.25 1429 1429, 0.34 2.60 GC 0.75 0.7 0.5 773 773 0.31 2.61 G 0.75 0.7 0.75 315 315 0.34 2.61 G 07 0.75 9 0.125 2042 1904 0.39 2.60 G 0.75 0.9 0.25 1431 1431 0.34 2.60 C_ 0.75 0.9 (10.5 773 773 0.31 2.61 G, 0.75 0.9 0.75 315 315 0.34 2.61 Cable 13.

I (lj I 019 01 I 1 016 17 I03 01Q.I I 2 2 I2 Q I3 Q24I Q25 I Q26I Q27 I 025 1 Q29 1 II [1.1 I fl') I /114 I f(J I 110 I fl/i fIt I flA~ Sim of Ch -r 5 50 4 191 451 60 1 GO 53T T T0 I .1 1 54 62 1 71 13] 61_ )1 72 16(4 J 63 1381 56 55 67 1138 78 1 79 1 73 107 1 79 1 77 1 IM 11Q Q Q4 I Q Q Q7 11Q8 (29 1 T01 (1-1 L Q214 Q151 16 1 17 Q8 Q19 (1Q20 21 22 Q3 Q-9 I Q5[Q8 Q27 QS I 91Q80 Q.I 0.3e 1.11 2.67 93 2.36 3.10 2.00 2.13 2.85 229 2.6.1 T2-5 2 7 B 2.15 2.7-1 .2'0 1 6 2.613 2.39 1.87 2.14 2.42 2.39 2.73 12.79 1.86 2.94 2.37 2.19 2.02 Q2 1.13 0.34 2.53 2-5O 2,2 0 2.(4 3.04 2.72 2.021 2-37 2' 7 -0 14 2-34 970 32.37 1.84 2.51 2.29 2.16 1.90 2.30 2.29 2.78 2.0 1.63 2.94 2.10 2.32 2.92 QJ 2.67 2.51 041 3. 65 2.28 2.2 279 2.1 30N '1 2 272'37 2(17 1'2 10 2i! 2.14 2.60 3.07 3.20 2.17 2.66 2.93 2.732 2.64 4.07 3.19 3A1 2.8 Q1 2.31 2.51 3.65 rI0- 2.48 418 33 VA) 2. 103 1 2.02 2.7G 3.1 243 3.211 3..12 3..53 5 3.26 1.74 1.07 2.94 3.28 2.15 2.30 2.37 2.12 2.13 1.93 3.1 Qj 2.36 2.25 2.2 2.48 0.39 3.07 2.41 2.64 2.71 2,18 2-4 2.2T 7 2.6l 2.18 1.2 2.48 1.96 2.39 2.51 2.01 2.19 1.90 2.51 2.12 1.90 2.09 3.05 2.10 1.92 2.30 Q. 3.10 2.94 2.83 41.27 3.07 0.37 3.14 2.56 45 2.72 3.90 3.59 2.0O 2.93 3.19 2-42 270 2.94 2.63 3.80 3.75 3.14 2.63 3.86 .1.75 2.50 4.95 3.91 3.88 0.05 Q7 2.911 3.01 2.79 2.44 3-1 0.41 2(2 3.09 13.01 2.23 1 3.7 2.322.2 2 2,1 2.57 2.47 2-73 2.92 2.46 2.47 2.701 2.73 2.76 3.8.5 2.90 2.85 1.34 Qr 2.7] 2.84 39 1 90 2.5(1 260 4011 253 3 j2 3 105 3.1 2.60 2.80 279 1.7 2 3.18 3.06 3.00 2.62 3.17 3.34 2.6,5 4.72 2.6 3.10 2.64 Q9 2.87 2.93 3.98 2.05 2.67 4.66 3.10 4.01 0.34 3.35 2.37 2.78 4.310 2.9 3.30 _oll 2.853.66 3.53 2.28 2,48 2.89 3.53 2.28 2.06 2.96 1.87 2.01 1.71 2.83 QIO 2.32 2.38 2.59 104 250 2.73 2.34 2.81 3.11 C 2.02 2.43 2 .27 131 2. 2.51 1.46 2.30 2.1 2.22 1.45 2.95 2.84 1.93 3.76 2.70 2.12 2.37 Q11 2.G0 2.1 3.2 1 2.02 2.32 3.88 3.05 3.140 237 2,96 0.37 2.8:3 3.44 2.03 2-N) 11 2.60 3.30 3.03 2.39 2.28 2.69 3.03 1.30 1.03 2.52 2.42 2.11 2.17 2.69 Q2l2 2.52 2.62 2.72 2.7.5 2.20 3.59 2.22 2.83 2.79 2. 4 2 0.40 2.88 2.79 2.02 2.29 2.5412.60 2.43 2.17 1.93 2.17 2.48 2.58 2.49 2.37 3.3 1.76 2.09 2.38 W4 .8 2.9 4 2.7 3.11 2.6(1 2 0n 215 188 II 911 2. 3.43 2.5. 3 1140 'IF6 56 2.38 2.65 2.37 2.3 1 3.23 3.12 2.87 2.34 3.06 3.24 2.69 4.4 3.14 3.3. 2.2 Q f .15 2.3-1 2.982.4.3 2.18 203 30- 7 2,0 231 2.04 2. 96 0.3 5 2.80 E60 18312.80 2.2 2.67 2.59 2.69 2.32 2.30 2.03 2.03 2.93 2.41 2.4 2.74 l3 2.71 .08 .91 3.20 1.72, 3.1)0 9.32 2(;0 3:10 229 2.98a 2.03 2.80 (22 (11 2.50 12.-13 2.04 2.45 2.41 1.31 2.04 2.01 2.62 2.51 3.64 2.74 2.78 216 Qt6 2.20 .12 2.2281 118 .30 3.17 2.20 2.38 2.65 2J S 1.39 21 2.35 1.33 2 .00 1.33 2.99 2.87 1.74 3.8q 2.99 2.89 2.37 (217 3. 60 1.83 2.76 2.48 1.97 2.,1 2.73 2.87 2.89 .22 .8i 2.0 266 18 2.50 9 0. 8 2.43 3.66 .34 3 1.63 2.63 2.5U 1.11 3.09 2.41 2.25 2.32 Q.18 2.63 2.51 2.13 3.55 2.30 2.64 2.56 1.97 3.84 2.5(0 3.2_ 2.65 236 2.88 2 15 2.38 2.43 0.39 2.33 3.15 3.06 2.71 2.33 3.10 3.25 2.17 4.38 2.97 3.17 2.95 Q19 2.37 2.262.631 3.22 2.5(0 2.72 241 2.77 3.34 '2.9 2.1 228 24 193 1-f -21 1 76 2.39 0.73 2.45 2.30 2.17 0.73 2.84 2.93 1.95 3.03 2.72 2.92 2.28 QPO 1.88 2.16 3.07 1.74 2.02 3.89 2.72 319 27- 236 2.39 2.17 3 24 2.66 2 1 2.36 2 -12 3.13 2.52 0.37 1.10 2.48 2.52 2.39 2.20 2.34 2.82 1.59 .1.31 2.64 QUI 2.1.1 1.90 3.25 1.97 2.19 3.i4 2.93 3.0T 24A 728 2.28 1.94 3.12 2.60 2.3 2.60 2.32 3.00 2.38 1.11 0.37 2.35 2.38 2.23 2.30 2.12 2.83 1.32 1.72 2.56 Q22 2.13 2.30 2.18 2.91 1.90 3.14 2.41 3.01 285 2.21 2.69 2.19 2.55 2.58 133 2.44 2,95) 2.73 2.20 2.48 2.39 0.37 2.20 2.38 2.13 2.33 3.45 2.72 2.62 2.13 Q2' 2.36 2.25 2.63 3.22 2.50 2.72 2.41 3 3 1.35 .6 2.51 2.28 2.44 95 1.20 110 2.38 0.73 2.44 2.29 2.19 0.73 2.84 2.93 1.05 8.92 2.72 2.92 2.29 (34 2.73 2.77 2.93 2.15 2.12 3.82 2.7 1 317 2. 27 2.95 1.30 2.583.07 2.30 2.03 3.02 23N 3.11 2.843 2.41 2.24 2.39 2.89 0.33 1.10 2.64 2.16 2.03 2.21 2.22 .o2 299 110. 0.3 7 531 71 35 1.4 2.14 2.24 2.02 221 Q25 .0 2.66 2.79 2.30 1.61 3.75 12.711 3.6 Q261 1.S7 1.02 2.65 12.3712.09 2.50 2.78 2.6 7 2.2962.04 4.086 2.13 3.05 1.95 3.86 4.70 CiO~~.19 I111 I99fl 91 I o 9~1 I a)0( I 9(1(1 O~tl I '.LII i -r.nll L.aa I L. m I e.Ls I r.ir~ I I huu I I ~.rY I I I 3 1932.55 2.37 2.69 2.03 2.51 1.7511.10 2.1812.031 2.35 2.13 2.33 2.

3.77 2.42 3.37 11.41i 2.92 3.6i 73T 2.1) I3.11 1 1.31 3.804 2. I .81 f8.43 3.

211016 V307 I227 2.41 2.26 Z.1CI 0.28 1 2.57 2.52 2.24 ~j I nnn 2 TO 2 12 176 M o 2.4111 2.Y5 2.991 1 2.401 1 2.98~ 2.74 l.UU I t1.oz 1 4.(11 T4 1 2.03n 2.24 -q -in 13 89298 3 49 2.21 2.32 3.06 1.93 1.91 3.89) 2.8.5 3.19 2.1055 21 ~l 3.33 2.48 12.78 12.88 2. .72g .17 .1 2.90122 T. 18 1 1 DR I10 IA) 322 10.32 1 1012.62 [123011.1o 0a35 2.4 i] 32 12.61112.63 10.47 1 £201289] 2.9 J 2.81 [2.28 1.1.04 1 1.3.5 2.: 2.411 Z-f-S -4-01) 1 Z-Oki 1 2,15 VLf I L U I I L IIl W II I I I I I WO 03/066678 PCT/AU03/00137 61 Table 14.

H: c.-helix B: residue in isolated beta-bridge E: extended strand, participates in beta ladder G: 3-helix (3/10 helix) I: 5 helix (Tr helix) T: hydrogen bonded turn S: bend U: no assignment WO 03/066678 PCT/AU03/00137 Table Cluster H B E G I T S U Cl 616 0.96 0.00 0.00 0.00 0.0 0.02 0.01 0.01 C2 682 0.01 0.03 0.56 0.01 0.00 0.03 0.05 0.32 C3 278 0.80 0.00 0.00 0.06 0.00 0.09 0.01 0.03 C.j 258 0.94 0.00 0.00 0.00 0.00 0.03 0.00 0.02 609 0.96 0.00 0.00 0.00 0.00 0.02 0.01 0.01 C6 1809 0.96 0.00 0.00 0.00 0.00 0.03 0.00 0.01 C7 255 0.92 0.00 0.00 0.00 0.00 0.06 0.01 0.01 CS 264 0.94 0.00 0.00 0.01 0.00 0,03 0.01 0.01 C9 300 0.97 0.00 0.00 0.00 0.00 0.02 0.00 0.01 CIO 274 0.95 0.00 0.00 0.00 0.00 0.03 0.00 0.01 C11 266 0.93 0.00 0.00 0.01 0.00 0.04 0.01 0.01 CI2 423 0.95 0.00 0.00 0.00 0.00 0.02 0.01 0.01 CIS 303 0.97 0.00 0.00 0.00 0.00 0.01 0.01 0.01 C14 1750 0.96 0.00 0,00 0.00 0.00 0.02 0.00 0.01 310 0.96 0.00 0.00 0.00 0.00 0.03 0.00 0.01 C16 304 0.96 0.00 0.00 0.00 0.00 0.02 0.00 0.01 017 476 0.96 0.00 0.00 0.00 0.00 0.02 0.01 0.01 0C18 1262 0.96 0.00 0.00 0.00 0.00 0.02 0.00 0.01 C19 345 0.96 0.00 0.00 0.00 0.00 0.02 0.01 0.01 340 0.97 0.00 0,00 0.00 0.00 0.02 0.00 0.01 C.1 944 0.96 0.00 0.00 0.00 0.00 0.02 0.01 0.01 C32 1667 0.97 0.00 0. 00 U.00 0.00( 0.02 0.00 0.01 C2S, 1648 0.97 0.00 0,00 0.00 0.00 0.02 0.00 0.01 C24 386 0.9 0.00 0.00 0.00 0.(00 0.02 0.01 0.01 882 0.96 0.00 0.00 0.00 0.00 0.02 0.00 0,01 428 0.960 0.00 0.00 0.00 0,00 0.02 0.00 0.01 855 0.96 I).D(I 0,100 0.00 0.02 U(.01) 0.01 C( 8 17:3 0.96 0.00 0. (00 (0.00 0. 0. 0.(2 .00 0.11 515 0.9 i .00 0.00 0i0 0.0:3 0.01 0.01 C(.0 1 I (0.9(7 oI.(l 0.I)0 0.00 0M)3 I(012 11.001) 0,01 WO 03/066678 PCT/AU03/00137 Table 16.

Cluster 1 Cj H B I E G I T IS U Cl C2 C3 C4 C6 C7 C8 09 Clai Cll C12 C13 014 C16 C17 C.21

CI

C26 C27 C28 C O

CIM

298 238 821 309 245 174 147 168 178 296 191 587 305 281 185 182 270 18.5 519 196 308 772 716 232 702 239 650 491 284 298 0.97 0.97 0.96 0.97 0.97 0.97 0.96 0.97 0.97 0.98 0.97 0.97 0.97 0.98 0.97 0.95 0.97 0.97 0.96 0.97 0.98 0.97 0.97 0.96 0.97 0.97 0.97 0.97 (1.97 0.97 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 (0.00 0.00 (0.00 0.00 0.00 0.00 0.00 0.00 0. 00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 (.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00) 0.00(( 0.00( 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0,00 0.00 0.00( oo00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.02 0.02 0.02 0.02 0.02 0.02 0.03 0.02 0.02 0.01 0.01 0.02 0.02 0.02 0.02 0.04 0,02 0.02 0.02 (0.01 0.02 0.02 0.02 0.02 O.02 0.02 0.02 0.01 0.02 0.01 0.00 0.00 0.01 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 1.001) 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 ().00 0.01 0.01 0.01 0.00 0.00 0.01 0.00 0.01 0.01 0.01 0.01 0.00 0.01 0.00 0.01 0.01 0.01 0.01 0.01 0.01()I 0.01 0.01 0.01 0.01 0.01 0.01 0.00 WO 03/066678 PCT/AU03/00137 Table 17.

Cluster Cj I H B E G I T S Cl C3 C4 06 C7 08 09 Cli 012 012 014 016 C17 C18 C19 (21 C.7 C26 0'29 101 113 98 112 91 91 115 109 93 95 200 184 154 113 109 97 115 188 270 113 125 122 125 137 256 198 156 140 143 0.97 0.97 0.98 0.97 0.98 0.98 0.98 0.98 0.97 0.96 0.98 0.98 0.98 0.97 0.98 0.97 0.97 0.98 0.97 0.96 0.98 0.97 0.98 0.97 0.98 0.98 0.98 0.98 0.98 0.9 c 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 .0.00 U.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0,00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.0(1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 U.00 0.00 0.00 0,00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0. 00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.02 0.02 0.01 0.02 0.01 0.01 0.01 0.01 0.02 0.03 0.01 0.02 0.02 0.01 0.02 0.02 0.03 0.02 0.02 0.02 0.01 0.02 0.01 0.02 0.01 0,01 0.01 0.01 0.01 0.01 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.01 0.00 0.01 0.01 0.01 0.01 0.00 0.01 0.01 0.01 0.01 0.00 0.00 0.01 0.01 0.01 0.01 0.00 0.00 0.01 0.01 0.01 0.01 0.01 0.00 0.00 0.01 0.01 0.00 0.01 WO 03/066678 PCT/AU03/00137 Table 18.

Cluster II CbI H B E G I 01 C2 C3 04 C6 07 C8 C9 ell CI2 018 C14 C16 C17 C18 C19 C21 C22 C23 C024 C26 C27 C28 C29 52 50 49 45 60 60 53 50 47 60 54 62 71 61 61 72 64 63 138 56 55 67 138 78 79 73 67 79 77 101 0.97 0.98 0.99 0.98 0.98 0.99 0.99 0.99 0.97 1.00 0.98 0.97 0.99 0.98 0.99 0.98 0.97 0.98 0.99 0.98 0.98 0.97 0.99 0.98 0.98 0.97 0.98 0.97 0.99 0.98 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00.

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 T S 0.01 0.00 0.01 0.00 0.01 0.00 0.01 0.00 0.02 0.00 0.01 0.00 0.01 0.00 0.01 0.00 0.02 0.01 0.00 0.00 0.02 0.00 0.02 0.01 0.01 0.00 0.02 0.00 0.00 0.00 0.01 0.00 0.03 0.00 0.01 0.00 0.01 0.00 0.01 0.00 0.01 0.00 0.01 0.01 0.01 0.00 0.01 0.01 0.02 0.00 0.02 0.00 0.01 0.00 0.02 0.01 0.01 0.00 0.01 0.00 u 0.01 0.01 0.01 0.01 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.01 0.00 0.01 0.01 0.01 0.00 0.01 0.01 0.01 0.00 0.01 0.01 0.00 0.01 0.01 0.01 0.01 0.00 WO 03/066678 PCT/AU03/00137 Table 19.

Motif size 4 C1 C2 C3 C4 C6 C7 C8 C9 Cl1 C12 CiS 014 C16 C07 C18 C19 021 C22 C23 C24 C26 C27 C28 C29 0.08 0.99 0.21 0.07 0.08 0.07 0.09 0.05 0.06 0.05 0.06 0.08 0.06 0.06 0.03 0.03 0.06 0.05 0.06 0.05 0.06 0.06 0.05 0.03 0.05 0.02 0.05 0.03 0.04 0.05 5 6 0.07 0.09 0.03 0.04 0.08 0.05 0.06 0.04 0.06 0.03 0.03 0.02 0.05 0.04 0.04 0.05 0.05 0.04 0.04 0.05 0.07 0.04 0.06 0.06 0.07 0.05 0.02 0.06 0.03 0.02 0.05 0.05 0.04 0.04 0.02 0.03 0.07 0.07 0.03 0.05 0.04 0.05 0.07 0.07 0.07 0.05 0.03 0.09 0.06 0.05 0.07 0.06 0.06 0.03 0.06 0.04 0.02 0.03 0.02 0.02 7 0.06 0.04 0.04 0.00 0.03 0.00 0.02 0.02 0.09 0.00 0.09 0,10 0.03 0.00 0.08 0.03 0.06 0.05 0.03 0.05 0.07 0.07 0.03 0.05 0.06 0.04 0.03 0.09 0.03 0.05 WO 03/066678 PCT/AU03/00137 Table Table 21.

Me~hud Siz uf moi~f TOL Scalcvc.) (l7j~l iteu Ici j' [A .Inauster RMSD ]Avg. IntercInster RMSD] o 4 0.75 0 J1625 1573 0.54 2.69 G' 4 0.5 0.2.5 6 67.1 3770 0.52 2.69 0.75 U 712 712 0.29 2.72 C 5 0.7 0.125 1264 1264 0.37 2.72 C 5 0.9 0.125 2298 2298 0.43 2.74 06 0.75 Cl 591 591 0.25 2.50 C6 0.7 0.125 680 680 0.26 2.50 G 0 0.9 0.125 771. 771 0.27 2.49 a 6 1.1 0.125 771 771 0.27 2.49 o 7 0.75 0 479 479 0.26 2.33 G 7 0.9 0.125 532 532 0.26 2.33 C; 7 1.1 0.125 532 532 0.26 2.33 Legend 0: One-pas3s algorAIthm G: Greedy algorithm Table 22.

111 I I Q4 1Q-I Q6I Q7 I7 a Qo I QH Q QIQ I 14 2 1 p, (1 7 Q18 Q19I QlI Q21 IQ2 Q C225 QI6 Q21 I228 I22 1071Q 511 (Iii i 3 1 20 20 18 9 1 55 21 2 1 1 2 201 110 474 311 21 50 19 1687 1:6 Ill 112 107 1649 7 216 228 :149 21 682 Q21 I (P I 04 Q, 1 I 727 I Q8 Q911 I QM(1 I QIJ I Q2 1 -2 I Q? I Q14 1 016 I 216 1 Q 17 1 QIS I Q119 I 201) 4111 I Q 2V 1 I Q 229 I 7 4 I 2P8 I Q26 I Q27 I Q.16 I 1 (230

~IS-

2,9:1 3.68 1- Aj 2.6 2.3 3 :.14 ]5 1.78 1.57 2.29 1:1 1.1)1 1.21) 1.AS 1.9 '1 1 1 1 2.99 .12 1.17 .27 4.1.1 3.2s 1.99 3.25 1 1.86 4.12 1.28 2. 1 2.53 1.27 13.22 5. 1.8.87 1 51 10 9.70 C02 2.77 3.24 11.17 41. 612.31. 2.81 2.2212.78 4.94 Z.42 4.07 4.8 13.61 3 4.85 .1.90 2.04 2'.80 1 2 1 l')1l .8 2.2 .6 e 2.16 .82 1.114 1.78I 1.70[ 1.8 1.892 2.74 2.82 [2.82 2.42 2.45, 3.71 1.2j [3.10 2.21 g 31 1 '.03 2.1)2 1.57 2.6 1.12 .1 1 II 2 .01 1.11 1.94 3.2 1.08 2.71 2.72 1.99 2.08 1.27 1(1 1.28 I 1 1 b '8 3-.112 2..2 1..11 2.11 1,.02 2.38 2.31) 2.17 1-i 4.1)2 2.88 2.98 2.97 2.1 2.X4 3.96 T .1111 2 AI1 2.28 .62.12 1.116 2.16 2.11 2.89) 2.021.86 2.2 2.610 2.6C1 2.1 2.!;12 3.54 1.9)1 2. 1 2701 2.55 3,80 1.:52 1.3 141 2.S 3.66. US .12 222 :1.1181 1.21 .21 .1.86 1.5T7 4.GJ01..3 3 :.831 IF 1.8 j 1 1.78i 1.701 1 ;.11:i1 2.1J4 I1.84 I 2.401 2.87 12.11j 2.76 1 1.61 1.61 2.317 I2.t17 I1.691 1.74 I1.61 3.06 2.1 B 4.89 .51 2.18 1 98 1.1) I .33 1.40(1 1.04 I 1.27 1 2.18 1 3.47 5.31 1 4.1.2 1 2.74 14.R) 1 2.76 1 5.31 4.27 1 1 3.49 j 4.7 I 4.24 I .21 LKI :1:111 4AU1 I 1.12 I 1,o[ 1 Q11 1.110 I1.21 1.1 I 1.49 3.19_10.25 4.41_ 1.67 ().6ui I I.S)I, .11 1.7,1 1 .5 F9 4.30 11 12 1 j 311 2. -5 3 l 1.3 11 3.54 13.3 1 3.48 1 1.21) 1 2.71 i 2.72 i 2.72 3.91.80 T_5.4 M.1ilI 3.42 1 1.1-, Q21 1.911 2.2 3 .281 2.1 13 LI 5-06 3.92 3.40 1.21 :.V7 1.61 5.32.

Q 1, 1 .263.64-2.3 191.196 2.81~ 1.511 1.17 0JI-j LO 2. 8C .71.14 2.2t 1.76, 4.0:1 Q11;~~ J3 2 7 2O)15 2.763 2.91) 1.645 1 42 Q)17 L318 3.272.'2 2.92 3.12 1.65 3.61 2.1 Q M 15 3.17 2.10 _t.52 2.56 2.10 2.12 3.41 S1v18 1.71 .12 5:8 Wu 1 38 2.32 2.92 2.21 1.12 1. 1.11 1.21 1.721 ().72 4.T III.21 1.2 1.71 1.57 J_.5 1,0 13.201 2.77 1.81I 2.7 1 1.06 2.0 j 1.68 1 .7 1.84 2.71) 1 1.77 2 .6612I 661 I 2)16i~ 2 .61 1.10 1 1,83 1 2.70 1 1.2 2.24 IM)21 1 08 4.2C :.A7 3.12 3.11 11.1- 2.74 4.241 3.37 2.17 3.951 2.1L5 4.24 1 1.121 2.68 2.67 3.451 .19 2.82 1.87 2.111 1.361 2.72 0,157 2.87 2.64 1.89 1.9611.7, [2.68 61.87 2.21 2.21 1.82 1. 82 2.52 1.11 1.78 i 1.f1 2.741 4.191 3.19 1.23 3.11 13.32 1 3.21 13.1 1 0.99 1 2.69 1 2.42 12.41 1 1.82 1 8 0.97 3.1 1 2.01 2.6V 15 11.,14 2.1U I LN 1 3.3,1 1 1.97 11.(1' .1 0.7 I 2.11 41.61 2.1 2. 12.17 11.73 1.7 2.99 2.12 11. i 2.29I 1.13.31 .68 1.6111.65 2 2.18 1 3.24 Q-1 .51 2.77 1.78 11.,1 7222 .B I112 N.6 1.1) .1 2.91 12.11 2.95 2.52 117 2.68 1.49 3.39 2.26 2. .5 1 2.21 2.12 1 I 2.971 1 i I I II' 31.10 1. 1m1 2.11ki I' I.-Ir 'i I 1 L' .91l 1. 1 i192.79 1 I315 2. 25 2.51 :1.111 1.12 j :1.111j I .0 1 ".J:U 4 .16 9.56 3.11 2.63 J. 1.118 1.78 1 3.2 1 22.012.8712.741.12 4.2 I .71 2.7 7 3.01 2.63 2.40.31.8 3 2.7S i.81j I .0011.791 2.55 i 2.601 0.66 1.7-,1 1.591 j.11 1.11 2.88 I .99 I 1.13 .1 1 I .ll 1.141 1. 1 1.79 1 (72 IF11 1-511 1 3.41 1 1.801 .1.74 1 1.76 2.21 1.86 1 2.18 1 1.73 1 2.06 1 2,02 2.021 0.614 I o.8n 1216 i3tr 1.-76 2.1f.97 1.L7 1( 1.7 11.6 1.65 0.357 1 2.34 2.539 1 2153 1 1.0.31 1.71 1 I(il Q1,2) 11 1.01 I 4.M2 1 31591 :1.1 3.29 32.12 I I.18 1 1.22 1 01.94 5 .16 1.J1 41 41 U 1.91 j1 1 .31 WO 03/066678 WO 03/66678PCT/AU03/00137 Table 23.

Cluster IW T~i H B T- G T~ s S cl 3 (1.67 0.08 0.00 0.00 0-00 0.08 0.00 0.17 012 20 0.00 0.00 (1.00 0.00 0.00 0-00 0.00 1.00 2(0 0.00 0.00 0.00) 0.00) 0.00 0.00 0.00 1.00 04/ 10 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 cu 38 0.03 0.00 0.00 0.01 0.00 0.03 0.01 0.02 55 0.01 0.01 0.83 0.00 0.00 0.00 0.03 0.12 C7 23 0.01 0.00 0.01 0.00 0.00 0.00 0.00 0.08 CS 3 0.25 0.00 0.00 0.17 0.00 0.25 0.00 0.33 C9 1 0.00 0.00 0.00 0.00 0.00 0.25 (1.00 0.75 040 10 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 C111 2 01.00 0.12 0.00l 0.0X) 0.00 0.00 (1.00 0.88 012 20 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 01.'I 110 0.00 0.03 0.50 0.00 0.00 0.08 0.07 0.30 014 474 0.01 0.02 0.63 0.01 0.00 0.02 0.04 0.26 31 0.01 0.05 0.04 0.02 0.00 0.02 0.02 0.85 0 11 26 0.01 (3.01 0.01i 0.00 (.10 0.0 0.01 0.95 017 50 0.95 0.00 0.00 0.00 0.00 0.03 0.01 0.02 018 99 0.01 0.00 0.81 0.00 0.00 0.00 0.03 0.10 019 1i6S7 0.03 0.03 0.319 0.02 0.00 0.06 0.11 0.35 CP0 136 0.01 0.01 0.74 0.00 0.00 0.01 0.04 0.10 (121 111 (3.00 0.00 0.80 0.00 0.00 0.00 0.02 0.16 022 102 0.01 (3.01 0.73 0.00 0.00 0.01 0.03 0.21 i 7 1.D0 0.010 0.82 0.00 0.00 0.01 (1.03 0.12 C) 1649 0.07 0.03 0.35 0.02 0.00 13.07 (1.10 0.35 9 (1.94 0.00 0.00) 0.00 0.00 0.04 0.11 C0?6 II236 0.02 (1.111 0.73 0.0 (1.00 (1.01 (0.02 (191 027 228 (0.02 (0.01 0.73 (0.01 (0.00( 0.01, (.013 (1.19 C'M~ 342 0.00 (1.02 (1.69 0.031 0.00 0.02 0.03 0.24 C19 290 OAR1( 0.11 01.71 11.00(t [1.01 (0.03 0.2:3 11i;,2 1(0.1 1 0.03) 1 (.r%50 M) 0.1 0.034 1)005 10:32 WO 03/066678 PCT/AU03/00137 Table 24.

Motif size 4 5 6 7 C1 C2 C3 C4 C6 C7 C8 09 Cll C12 C13 014 C16 C17 C18 019 C21 C22 C23 C24 C26 C27 C28 C29 0.33 1.00 1.00 1.00 0.16 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.99 0.99 1.00 1.00 0.12 1.00 0.96 1.00 1.00 0.99 1.00 0.92 0.11 0.99 0.99 1.00 0.99 0.99 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.17 1.00 0.08 1.00 1.00 1.00 0.08 0.99 1.00 0.13 1.00 1.00 0.18 0.99 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.15 1:00 1.00 1.00 1.00 1.00 1.00 0.13 0.11 1.00 0.14 0.14 0.14 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 WO 03/066678 PCT/AU03/00137 Table Cluster Cji\ H B E G I |T I S IU C2 C73 C4 C6 C7 C8 C9

CII

CIS

C13 C14 C6 C17 C19 C21 C22 C23 cp- C26 C27 C28 C29 0.97 0.96 0.97 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0. 00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0. 00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0,00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00- 0.00 0.00 0.00 0.02 0.02 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.00 0.01 0.00 0,00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.01 0.01 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 WO 03/066678 PCTIATJO3/00137 Table 26.

Type of motif C -l S 1. 45. 3.,2 2. 98 22.0 24.4 43.0 25.2 4.3 42.9 26.7 39.71 23.5 23 39.0 23.9 4 41.0 19.9 3.4 42.3 19.4 35,6 .40.1 67.8 35.3 40.3 69.0 33.

U 3. 40.2 64.9 325 41.5 64.6 o34.6 38.2 61.9 35.1 36.9 62.2 4 3.6 365 9.7 304 38.3 59.9 ~49.4 26.0 67.5 50.3 29.6 68.7 ~U 46,5 23.9 66.1 45.5 23.8 67.2 Q) 50. 21.1 68.6 51.5 20.0 68.7 4 47.0 20.9 70.2 47.2 20.9 71.7 S 9.8 2.9 30.7 10.4 2.3 31.8 U 6.1 2.0 30.7 5.1 2.4 31.7 8 8.4 2.6 25.8 9.6 2.6 26.2 x D 7.1 -1.0 26.7 6.9 -1.5 28.1 7.1 -1.9 21.6 8.2 -2.4 225 14.2 1. 492 13.9 11.1 50.6 U 14.1 0. 4.2 15.1__10.1 46.2 U 109 6. 44. 10. 9.946.0 U 11.6 6. 02 11.8 6.0 41.5 7.9 6.8. 39.8 6.9 7.5 40.7 (19.61 4.6 35.6 9. C401 36.61 23.8 10.2! 80.4 22.4 10.7 60.1 (n 23.3 6.'619 22,6 5.5 61.1 o H246 8.8] 6.3 23.1 6. 6 .0 Ii 26.1 5.41 9722.643 6.

26.0 6.81 72.0 25.0 5.9 71.4 28A4 8.5 76.3 27.0 9.0 75.9 282 4. 9 77.4. V7. 3.8- 76.6 8 0.5 4.4 21.7 81.1 4.4 _20.7 M 7&2 7.2[ 22.7 78.1 7.,3 24.1 (9 73.6 11. 8 21.9 _74.0 12.2 23.0 70.4 13. 1 20.3 70.0 1 2.6 1 9.3 U8 14.8 -1.5 24.2 13. -0.7 24.1 U 15.5 -2.5 20.6 15.9 -1.6 19.4 8 18.8 21.7 20.0 -3.4 22.3 01 1 -1.2 19.9 14.9 -12.3 168 I15.8 -18.6 20. 1 15.4 -17.9 1.

_5.1 -10.3 1.7. -10.1 0.9 02 3.2 -7.1 2.3 3,6 -6.3 U 2 8 -06 .0 -3.7 0j 40. 16 4.2 -1.51 32. 0 4.3 2.6 0.

;a3 .91 -1.71 1.1 9.11 -2.31 -2.9 30.2 -0.9 31.4 -1.2 6 -4:9 33.4 -0.3 -5.3 33.7 1.

U -4.1 36J.7 2.1i -5.3 37.5 -2.7 0 -2.5 42.3 1.4 -1.8 42.6 2.7 3 486 -4 -42 _46.5 '-0.1 01 481 3 4. 509 1. 21611J.

Claims (21)

1. A computer-implemented method of producing a descriptor of protein Ssurface shape including the steps of: identifying a three-dimensional surface shape of each of a plurality of proteins, wherein each three-dimensional surface shape is represented by a side-chain location and orientation of two or more amino acids of each of said plurality of proteins; and (ii) creating one or more said descriptors wherein each said descriptor Srepresents a common surface shape derived from respective said three- dimensional surface shapes of three or more non-homologous proteins of said plurality of proteins.

2. The method of Claim 1, wherein the or each said descriptor represents a common location and orientation of respective side chains of three or more amino acids of each of said three or more non-homologous proteins.

3. The method of any preceding claim, wherein the location and orientation of each said amino acid side chain is in three-dimensional (3D) space.

4. The method of Claim 3, wherein each amino acid side chain used to produce said descriptor is simplified as a Ccc-C3 vector. The method of any preceding claim, wherein each said descriptor represents a common charged surface region of each said protein.

6. The method of Claim 5 wherein each charged surface region is represented by at least four grid points.

7. The method of Claim 6, wherein respective said grid points are 0.2 to angstrom apart in three dimensional (3D) space.

8. The method of Claim 7, wherein respective said grid points are 0.5-1.5 angstrom apart in three dimensional (3D) space.

9. .The method of any preceding claim, wherein said three-dimensional surface shape is of at least part of a structural feature of each of said proteins. The method of Claim 9, wherein said structural feature is, or comprises, a

13-turn, a loop or a contact surface. 11. The method of Claim 9, wherein said structural feature is, or comprises, a loop or a contact surface. 00 12. The method of Claim 11, wherein the contact surface comprises one or more discontinuous and/or continuous surfaces. 13. The method of Claim 10, wherein said descriptor represents side-chain location and orientation four 1-turn or loop amino acids.

14. The method of Claim 11, wherein said descriptor represents side-chain location and orientation of at least three amino acids of a contact surface. The method of Claim 14, wherein said descriptor represents side-chain location and orientation of four, five, six or seven amino acid side-chains of a contact surface.

16. The method of any preceding claim, wherein the common surface shape is N, an averaged surface shape.

17. The method of Claim 16, wherein the averaged surface shape is a mean, median or mode surface shape.

18. A computer-implemented method of identifying one or more molecules having a common three-dimensional protein surface shape, said method including the steps of: creating a query using one or more descriptors produced according to any one of Claims 1-17; and (ii) using said query to search a database and thereby identify one or more entries in said database that correspond to one or more molecules that each match said descriptor.

19. A computer-implemented method of creating a library of molecules including the steps of: searching a database to identify one or more entries corresponding to one or more molecules that each match a descriptor produced according to the method of any one of Claims 1-17; and. (ii) using at least one of the one or more molecules identified at step (i) to create a library of molecules. The method of Claim 19, wherein said library of molecules is a virtual library.

21. The method of Claim 19 said library of molecules is a synthetic chemical library.

22. A method of engineering one or more molecules including the steps of: 00 creating one or more descriptors according to the method of any one of Claims 1-17; and (-i (ii) engineering one or more molecules that respectively comprise one or more structural features according to the or each descriptor in

23. The method of Claim 22, wherein the one or more molecules comprises a protein.

24. The method of Claim 22, wherein the one or more molecules comprises a small organic molecule. A computer-implemented method of producing a descriptor of protein (Ni surface shape substantially as described herein with reference to the accompanying drawings.

26. A computer-implemented method of identifying one or more molecules having a common three-dimensional protein surface shape substantially as described herein with reference to the accompanying Examples, Tables and Figures.

27. A computer-implemented method of creating a library of molecules substantially as described herein with reference to the accompanying Examples, Tables and Figures.

28. A method of engineering one or more molecules substantially as described herein with reference to the accompanying Examples, Tables and Figures.