JP2003519201A - Highly efficient mapping of molecular variants to functional properties - Google Patents

Abstract

  (57) [Summary] A method for selectively changing a multivariable molecular synthesis for optimizing the functional characteristics of a reaction product, wherein a functional characteristic output obtained by reacting all primary library input variable combinations to the multivariable synthesis is obtained. Constructing a primary library of data and measuring the functional properties for all reaction products, wherein the primary library input variable combination comprises all selected values for the variables employed one at a time; Ordering the values for each input variable according to the effect of the input variables on functional property optimization; and a low frequency sampled secondary library input variable combination employing two variables at a time from the ordered input variables. A secondary library of functional characteristic output data obtained by reacting the set of Process and measuring; step and interpolating between the functional characteristics output data to optimize the functional characteristics; process comprising.

Description

Detailed Description of the Invention

(Cross Reference of Related Applications) This application is based on US Patent Law Section 119 (e), priority of US provisional patent application No. 60 / 174,225 filed on January 3, 2000. Claim the benefits of. The disclosure of this patent application is incorporated herein by reference.

BACKGROUND OF THE INVENTION In chemical technology, many laboratory experiments, environmental and industrial processes, and model exercises are characterized by a large number of input variables. The general purpose in such cases is often to keep the effect of the high dimensional input variable space on the observable system behavior as complete as possible in order to better understand the phenomena, often with optimization in mind or simply related. To scrutinize. An important issue in performing these scrutiny is the number of experiments or model variants needed to efficiently learn the input / output behavior of the system.

In conducting a chemical / physical system experiment or modeling in which there are a large number of input variables available for the modification, the modification of the input variables may be done with certain design strategies in mind, or , Can occur randomly due to the natural changes in the uncontrolled inputs. In all situations, the common goal is to scrutinize the effects of the input variable space on one or more system observables in question and do as much processing as possible. Such exercises perform optimizations to physically understand the role of input variables, or in many cases, ultimately to achieve one or more desired physical objectives by special selection of input variables. Therefore, it can be implemented.

Molecular materials (ie samples of a single type of molecule) encompass many applications such as muteins and drugs. The i-th variable for the molecular substance may be the i-th site for chemical functionalization on the reference molecular structure. In the case of proteins undergoing amino acid mutations, the total number of variables (ie mutation sites) can be very large and the variables x _i associated with backbone site i can take values up to 20 beyond the natural amino acids. . Drug molecules, in contrast, are usually of a modest size generated by the functionalization of minority sites on a reference chemical scaffold. In the case of drugs, the i-th variable can take many values, because the chemical moiety, which is quite arbitrary for substitution on the appropriate molecular scaffold.
This is because a set of l moities) can be considered.

Molecular substances are inherently different from those in mixture preparations, because the input variables for the molecular parts (eg methyl, ethyl, chloro, etc.) are discrete while the input variables for the mixture are This is because the component mole fraction can take a continuous value. Mixed materials from a large set of possible molecular species have both discrete and continuous variables. All of these substances have the problem of being characterized by either a large number of input variables or a large number of variable values, thus leading to high throughput synthesis in attempting to handle the possible explosive number of material samples that can be produced. /
There is great interest in screening techniques.

A property that is common to all of the above examples and many other examples is the large number of variables that can naturally occur to describe the input. The concept of "many" in this context is based on the individual application, and in particular on the difficulty in properly observing or calculating the system output corresponding to any single detail of all input variables. . Similarly, the drug search involves a small number of variables (ie, sites for functionalization on the molecular scaffold), but the number of moiety values for each of these variables can be very large, above 10 ^2. obtain. In this case, it is easy to generate one possible drug molecule, but it is difficult to realize all relevant possibilities. In another problem, the number of variables involved is inherently large, one example being the case where there are hundreds or more discretized input variables because the inputs are functions and good resolution is required. Occurs.

Therefore, due to the exponential relationship between the number of molecular variants in the system to be scrutinized and the number of possible variable combinations in the system, the discovery of molecular substances with the desired properties is still a great burden. is there. Combinatorial chemistry has been attempted to address this issue with random or quasi-random synthesis techniques. Alternatively, modeled procedures have been attempted to achieve the guidelines by design.
However, none of these technologies have reached full promise and functionality. Therefore, there is still a need for ways to structure and interpret data so that scrutiny of multivariable reaction systems can be further utilized.

SUMMARY OF THE INVENTION This need is met by the present invention. When a reasonable initial number of experiments were used with the technique employed by the present invention and further experiments were guided in detail, in a very efficient manner without reacting all possible combinations of variables in the system. It has been found that multivariable reaction systems can be scrutinized. The present invention decomposes the effect of each variable into a hierarchy of cooperating terms and recognizes that the contribution of any particular variable can be directly assessed from laboratory observations based on rational ordering of molecular variables. Show that the number of molecular synthesis experiments required to carry out a complete study of a multivariable reaction system increases at most with a polynomial according to the number of sites, and under certain conditions it actually It incorporates the discovery that it will be immutable.

Therefore, according to one aspect of the present invention, there is provided a method for selectively altering a multivariable molecular reaction that optimizes the functional properties of a reaction product, wherein A step of constructing a primary library of functional characteristic output data obtained by reacting a combination of library input variables, and measuring the functional characteristic for all reaction products, wherein the primary library input variable combination is 1 at a time. Including all selected values for each of the adopted variables, other variable values are held constant or randomized, and the input variables for the functional characteristic optimization based on the above primary library of the functional characteristic output data. Ordering the values for each input variable according to the effect, and using a second-order library input sampled infrequently from the ordered input variables A step of constructing a secondary library of functional characteristic output data obtained by reacting a set of force variable combinations and measuring the functional characteristic of all reaction products, wherein the secondary library input variable A combination is two variables that are combined at once from the infrequent sampling of the ordered input variables, while other variable values, if any, are held constant or randomized. And a step of interpolating between the functional characteristic output data for optimizing the functional characteristic.

The secondary library described above uses ordered infrequent sampling of ordered variable values, fully random infrequent sampling of ordered variable values by a combination of two sampling techniques or other infrequent sampling. It can be constructed by reacting by a technique. Which sampling technique is used depends on the number of variables and the laboratory synthesis technique.

The method of the present invention uses a conventional High Dimensional Model Representation (High Dimensiona).
1 is a modification and improvement of the 1 Model Representation (HDMR) algorithm. HDMR directly guides laboratory synthetic studies by adopting a limiting set of constitutive principles, resulting in rapid identification of compounds with desired functional properties. The present invention improves the HDMR technique by suggesting a second step in a modest synthetic study using ordered sampling of synthetic reaction variables, and finally,
This will lead to a means to quantitatively evaluate the functional properties of molecules, including the entire range of possibilities, including those that have not been synthesized.

The method of the invention therefore first carries out an initial set of syntheses containing all the values selected for each reaction variable and observation of the functional influence of each reaction variable, and then on the reaction products. A further selective set of ordered or randomly sampled molecules is performed by performing an algorithmic analysis of the above results in which the values of each variable are ordered according to their effect on the optimization of the functional property in question. Is suggested. The global information obtained from these syntheses and their functional influences is then reconstructed into a high-dimensional model representation in order to interpolate the functional properties over the entire space of molecular possibilities. As an example of possible labor savings, cases involving millions or more of molecular possibilities can be quantitatively studied with as few as 200 molecules wisely synthesized.

Other features of the invention are pointed out in the following description and claims that disclose the principles of the invention and the best mode presently contemplated for carrying it out.

A more complete appreciation of the present invention and many other contemplated advantages can be readily appreciated by reference to the detailed description of the invention in light of the accompanying drawings.

Detailed Description of the Preferred Embodiments FIG. 1 shows an application program embodying the methods of the present invention to discover molecular substances such as reference molecular structures under study for drug action.
In step 10, a primary library of functional characteristic output data is constructed by means essentially conventional to HDMR. For multivariable reactions involving the variables a, b, c, ... Z, the values for each variable are a ₁ , a ₂ , a ₃ ... a _N ; b ₁ , b ₂ ,
It is selected such as b _{3 ···} b _N. Input variable combinations are prepared by adopting one variable at a time, while other variables of the reaction are held constant or randomized. Therefore, while the combinations are prepared for the variables a ₁ , a ₂ , a ₃ ... A _N , the variables b to z are held constant or randomized for all variables of the reaction. One of the preferred embodiments of the present invention, which represents an improvement over the conventional HDMR, is to sample the entire space completely randomly, with the only requirement that all of the selection variable values be included in each sampled combination. Forming variable combinations. No matter how the input variable combinations are formed,
The combination considerably reduces the total number of combinations to be reacted in order to probe all possible combinations.

A primary library is constructed by reacting each combination and then measuring the functional property or properties of interest for all reaction products. Its purpose is
All selected values are reacted first for each variable adopted one at a time, while other variable values are kept constant or randomized to determine the effect that the selected value of each variable has on the functional characteristic in question. To obtain a one-dimensional sampling that provides a measurement, for each variable across a multidimensional space.

For molecules under investigation for drug action, the variable set is the basic molecular structure or skeleton with one or more chemical scaffolds, ie, one or more sites for chemical functionalization, as well as the function of each site. It may include various chemical moieties that are selected for chemistry (eg methyl, ethyl, chloro, etc.). The variables also include structural alterations and spatial features of the skeleton. Functional properties to be optimized also include drug action, side effect minimization, bioavailability, improved product yield, and other properties essential for drug discovery of interest.

Thus, for the purposes of the present invention, the “value” of a variable does not necessarily refer to an empirical amount, but to the identity of the various alternatives under consideration. If the variable a represents a chemical skeleton and the part should be changed at each site of chemical functionalization on the skeleton, then a ₁ , a ₂ , a ₃ ... Denote the skeleton structure to be studied. Represent If the variable b represents the part to be modified at the first chemical functionalization site on the framework, then b ₁ , b ₂ ,
b ₃ ... B _N represent the various moieties selected to be tested at their functionalization sites, such as chloro, fluoro, methyl, ethyl and the like. Other variables and values are assigned to the remaining chemical functionalization sites.

The present invention is not limited to the study of molecules for drug action, but is applicable to general molecular substances. In addition to molecular use as a drug, non-limiting examples of problematic molecular uses include molecules that have optical, magnetic, or electrical effects. As in the case of the molecules studied for drug action, the variable set consists of one or more chemical scaffolds, or basic molecular structures or scaffolds with one or more sites for chemical functionalization, as well as each site. It includes various chemical moieties that are selected for functionalization. The present invention is also generalized to polymeric materials consisting of individual monomer moieties, each of which comprises a chemical framework having one or more chemically functionalized sites that may be modified for the purposes of the invention. Is applicable. The functional property to be evaluated is the functional property of the polymer.

Another example of a multivariable system for use with the methods of the invention is a candidate protein for site-directed mutagenesis. Each amino acid within the protein sequence represents a possible independent variable selected from each amino acid employed in protein synthesis. Functional properties to be optimized are protein properties such as folding stability, ligand binding affinity.

Each variable can be derived by an initial combinatorial chemistry screen for possible molecular structures or from lead compounds as drugs. For example, natural compounds found to have therapeutic properties provide chemical scaffolds to be optimized as well as related scaffolds to be studied, as well as treatments for efficacy, efficacy, safety, bioavailability, etc. For the purpose of optimizing the effect, the chemical moieties to be altered at the chemical functionalization site of each scaffold are selected. Another example of a drug lead compound is a protein that has been found to have a ligand binding affinity that produces a therapeutic effect, which comprises one or more sequences with the aim of improving the ligand binding ability to enhance the therapeutic effect. Amino acid sequences for site-directed mutations at positions are provided.

Molecular material is an example of a system where the input variables have discrete values. Other systems to which the present invention is applicable may have variables with continuous values such as component mole fractions of chemical mixtures. Reaction conditions such as temperature, pressure, reaction time, etc. may also introduce continuous-valued input variables to variable combinations that would otherwise be limited to variables with finite and discrete values. However, continuous-valued variables require the selection of data points that provide a complete sampling of the entire variable continuum. Whether or not each variable in a multivariable system is continuous or discrete, the goal is to react all selected values first to each adopted one at a time, while Is to be kept constant or randomized.

In step 20, the primary library output data is used to order the values for each variable such that the property under consideration changes monotonically with the variable value. In other words, the functional characteristics can influence the ordering of the selected values, and the selected values for each variable are ranked in terms of their functional characteristics as observers, so that the value that produces the best optimal functional characteristic has the greatest importance. Assigned, and so on until the value that produces the lowest optimal functional characteristic. The purpose is to identify a "natural" order of values for each variable, where "natural" is defined as a rational ordering of variable values based on actual experimentation on functional properties.

Steps 10 and 20 may be applied to one or more functional properties, each functional property having its own natural order of variables. One purpose may be to find a reaction product with an optimal combination of two or more functional properties, in which case step 10 performs a measurement of two or more properties and step 20 essentially comprises Each variable value will be scored based on the results achieved for the optimization for each property. Each functional property is weighted differently for the purpose of scoring variable values depending on the importance of the functional property to the final purpose sought for the optimized reaction product.

In step 25, the ordered variables are evaluated to ensure that the output data for each variable behaves more or less regularly over the range of sampled values. For the purposes of the present invention, "regular" is defined as the state in which the difference in the functional characteristics between the closest variable values is "smooth", ie as small as possible. If not, the method proceeds to step 27, where the primary library is refined (
refined). This refinement can be accomplished in several ways, but the goal is to expand the set of primary library input variable combinations. One way to do this is to repeat step 10 and again adopt one variable at a time while holding or randomizing the values of the other variables to form a new input variable combination to produce another output. To find a set of data. Another way to do this is to increase the number of variable values. This involves probing of another chemical scaffold, or evaluation of additional moieties at chemical functionalization sites, using examples of molecular structures being studied for drug action. The method is then step 10 (measurement of the combination response and functional properties), step 20 (ordering of variable values from each functional property), and step 25 (ordering for smoothness regularity). The evaluation of variable values) is repeated, and step 27 is followed again if necessary.

Another approach to increasing the primary library input variable combinations is to find complete or partial secondary output data and introduce this into the reordering of the variable values. In other words, by adopting two variable values at a time and holding or randomizing the other variable values constant, some or all of the above primary library input variable combinations are reconstructed to make possible variable combinations. A more complete sample of and better assessment of the cooperation between each variable is needed. Steps 10, 20 and 25
Is repeated again, and step 27 follows again, if necessary. Regardless of which option is selected, when steps 10 and 20 are repeated, the values of at least one of the variables are reordered. This is because the increased data provides a new insight into the rational hierarchy of variable values in terms of functional properties, which results in a "smoother" or more regular ordering of the data. .

If step 25 confirms that the ordered variable values are as regular as possible, then the method proceeds to step 30 where the infrequently sampled secondary library input from the ordered variable values in step 20. A secondary library of functional characteristic output data is obtained by reacting a set of variable combinations. For the purposes of the present invention
"Infrequent" sampling is defined as a modest partial sampling suggested by observing the effect that values have on each variable with respect to functional property optimization, which may include reaction products that are not synthesized. Is effective in allowing a quantitative assessment of functional properties for the reaction products over the entire space of the low frequency sampling to achieve all possible variable combinations to reach the reaction product with the optimal functional properties. An economic measure is shown in comparison with sampling.

The above low frequency sampling may be carried out in several ways, the purpose of which is to seek a rational assessment of the functional characteristic performance over the entire space of possibilities. For example, ordered sampling may be performed, where ordered variable values are constructed with multidimensional axes or arrays and sampled periodically,
Every 5th, 10th, 20th or 100th variable combination is sampled. Therefore, increasing the number of combinations sampled increases the resolution required over the variable space. Alternatively, a completely random sampling of the variable space may be performed. Ordered or random sampling in this way can be performed uniformly or non-uniformly over the entire space of possibilities. Non-uniform sampling is performed iteratively until the region of optimal functional characteristic performance is identified.

Low-frequency sampling techniques for secondary library construction are otherwise HD
A detailed description is unnecessary because it is inherently conventional for MR mapping and is easily adopted by those skilled in the art. However, the main contribution of the present invention is not in the low-frequency sampling technique, but in reasonably and naturally ordering the variable values from the viewpoint of functional characteristics, and therefore the first low-frequency sampling is possible. Is done. Those of ordinary skill in the art, with reference to the present specification, should be able to modify existing HDMR software algorithms without undue effort to achieve the objectives described herein if the concept is understood. Is clear.

Conventional low frequency sampling techniques, including those known as cut HDMR and RS-HDMR, are described by Rabits et al. In “General Founda”.
conditions of High Dimensional Model Repr
"Essentation", J. Math. Chem., 25, 197-233 (
1999). Truncated HDMR employs regular low-frequency sampling around a reference point called the cut center, while RS-HDMR determines a low-frequency Monte Carlo sampling expansion function over a multidimensional space.

The secondary library construction described above is otherwise conventional for HDMR. Being a quadratic library, the input variable combinations are constructed by adopting two variable values at once with alternative variable choices based on low frequency sampling techniques.

Step 40 represents a functional characteristic measuring step. Functional characterization is performed on the reaction products of all secondary library input variable combinations to obtain a complete set of output data for the secondary library.

The method proceeds to step 50, where output data from the primary and secondary libraries are interpolated to identify combinations of input variable values to produce reaction products with optimal functional properties. Again, this is essentially a conventional step for HDMR. Often, the identified input variable combinations have not previously been reacted, in which case the method proceeds to step 60 for the reaction of the identified variable combinations to produce a reaction product with optimal functional properties. move on. Step 60 is
The result is verified to determine if the result based on the interpolation of the functional characteristic output data is acceptable in its current form. If the results are acceptable, the method is complete. This is because the reaction products with the optimal functional properties in question have been identified. If the result is unacceptable, there are several options.

The first is to proceed to step 70 to refine the secondary library. This essentially increases the output data resolution for the secondary library by performing even less frequent sampling to find more data points for interpolation. For example, if ordered sampling of every twelfth combination is being performed, then every tenth combination may be sampled for additional data points. If the input variable combinations are randomly sampled, then additional random sampling may be performed to find additional data points. Also, if the random sampling was non-uniform, other regions in the space of possibilities could be randomly sampled non-uniformly until the region with the optimal functional characteristic performance is identified.

The secondary library can also be refined by performing correlated input variable combinations based on the information in hand. If each variable is represented as a coordinate axis in a multidimensional space and the selected variable values are located on the coordinate axis according to the natural order of the selected variable values recognized by the functional property, the correlation combination is essentially By matching the recognized natural variable values for two or more variables as closely as possible, we represent the rotation of the coordinate axes into a region that is in agreement with the optimal functional characteristic performance. Essentially this is a multi-dimensional cut across the space of possibilities to collect additional data points by scrutinizing the area of possible optimal functional properties performance.
l cut).

The present invention contemplates iterative iterations of steps 30, 40, 50, 60 and 70 until sufficient output data has been determined for the above interpolation step to identify reaction products with optimal functional properties. However, at any point in the process, regardless of whether steps 30-70 were performed once or multiple times, step 20
It is desirable to repeat. That is, the secondary library output data is an ordering that cannot be identified by the information held at the time when the first ordering is performed, and a more reasonable ordering with respect to natural variable values from the viewpoint of functional characteristics. Can be represented. The method may repeat the steps by returning directly from step 30-70 to step 20, or, if step 20 is repeated, first using the gripping information to determine additional data points. The process may be repeated by returning to step 10 to build another set of primary library output data.

Steps 30-70 may then be repeated, if any, with additional infrequent sampling based on the variable value hierarchy generated by the iteration of step 20.
The iterative iteration of steps 30-70 may be continued, including returning to the step for reassessment of the reasonable ordering of the natural variable values until the reaction product with optimal functional properties is identified.

One result is that even if reaction products with optimal functional properties were identified over the entire space of possibilities defined by the values chosen for each variable, the optimal value was the origin of the above study. It may still be insufficient for the target values set in. Taking the molecular structure being studied for drug action as an example, the compounds identified as having the highest potency in all possible combinations, for example, still have too low potency to be further studied as candidate drugs. Sometimes it's not worth it. Perhaps this is the result of the selected values for one or more variables, then the solution proceeds to step 90 of the present invention, where each set of variable values is expanded and the above method repeated from the beginning. It is to be.

Thus, in step 90, an option is provided to expand all or some of each set of variable values and then repeat the method of the present invention. For example, additional chemical scaffolds may be selected for study, or additional chemical moieties may be selected for one or more chemical functionalization sites. The purpose is to expand the space of possibilities in looking for areas of functional property optimization in the intended goals for functional properties.

Another result is that, due to the cooperativity of cubic and possibly quartic, interpolation cannot be done accurately over the entire space of possibilities based solely on primary and secondary library output data. Can be. Since chemical systems are usually defined by low-order multivariable cooperativity, in most cases primary and secondary library output data are optimal by seeking enough information to allow interpolation over the space of possibilities. Sufficient to identify reaction products with functional properties. However, since the possibility of interdependence of variables increases especially in the case of a system containing a large number of variables, it is necessary to scrutinize the combinations of tertiary and quaternary variable values and output data derived from them when possible. Becomes

This is shown in FIG. 1 at step 100, which is a third and even higher order of functional characteristic output data derived from the response of the input variable combinations that were sampled infrequently from a library of previous output data. Check the construction of the following library. That is, the ternary library employs low frequency sampling, which employs three variables at a time from the output data of the secondary library, while the quaternary library employs low frequency sampling, which employs four variables at a time from the output data of the tertiary library. It is based on sampling, and so on. At any point in time, the output data can be interpolated according to the method of the present invention to identify reaction products with optimal functional properties.

Certain embodiments of the method of the present invention are illustrated with reference to the following description of applications for site-specific protein mutations, which are intended to limit the scope of the invention as defined by the claims. Should not be interpreted.

HDMR Reordering Applied to Site-Specific Protein Mutations Most of the protein mutations to date have been guided by intuition regarding protein chemistry and the physical / chemical properties of amino acids. Including mutations,
It was highly selective. Thus, such studies rarely systematically analyzed the large set of single and multiple mutants required to assess the effectiveness of HDMR reordering. Certain systematic data are available for the gene V protein of bacteriophage fl, a homogeneous dimeric single-stranded DNA binding protein (Sandberg et al., Proc. Natl. Acad. Sci.
． USA, 90, 8367-8371 (1993); Sandberg et al., Bi.
ochem. , 34, 11970-11978 (1995)). The two contact positions Val35 and Ile47 within the hydrophobic core of the protein were partially randomized, making eight substitutions at each position, including many of the corresponding double mutants. The effects of the mutants were evaluated semi-quantitatively by their temperature sensitivity of their phage growth phenotype, and all single mutants and 18 of the double mutants were purified to yield folding stability and It has been characterized in terms of both DNA binding affinity.

2A, 2B, 3A, and 3B, Sandberg et al. (1995).
Using the stability and binding data of, a relatively sparse but still informative matrix can be plotted. Reordering each variable based on laboratory data using the wild-type protein as the cleavage center illustrates that a regular pattern can be identified. 2A and 3A use the original ordering of residue substitutions provided by Sandberg et al. Despite the fact that the effects of single-site substitutions were relatively mild and most of the pairwise substitutions provided additive effects, no pattern was evident in the plots of FIGS. 2A and 3A. . In contrast, FIGS. 2B and 3B show that reordering the response data to produce a monotonic behavior along each axis results in a relatively smooth response across space. Not all spatial response surfaces are monotonic in the general case, but almost so here, and probably when significant non-additive effects are present. However, only the response surface needs to be reasonably regular to be able to use interpolation and HDMR analysis over the whole space. An important consideration in such studies, especially in the case of mutations at many sites, is the accuracy of the data. Although HDMR provides a systematic means of performing and interpolating low frequency sampling over the entire space of possible mutations, the accuracy of the observed data is marginal. One conclusion from HDMR is that a few mutations should provide an assessment for protein functional properties across space, subject to attention to quality of observation.

As is apparent in the gene V protein example, each functional property may have a different monotonic reordering with respect to variables, thus making the functional property itself unique to the relative significance of each term in the HDMR expansion. Pattern. Similarly, the resulting regularity of the response surface after reordering each variable means that there is a pattern in the properties (side chain type) of each variable. The interpretation of the data suggested by this method is likely to provide a more comprehensive understanding of the properties of individual residues that are relevant to the function or properties of various molecules.

The above variable reordering method followed by HDMR analysis is not limited to the study of protein mutations, but may include other types of molecules (eg drugs and genomes).
And their associated observable properties. Moreover, this reordering method does not require a priori understanding of the relationship between components and observational properties, instead:
Observational characteristics may define the relationship. As an example, FIGS. 4A and 4B show Sand
It shows a semi-quantitative temperature sensitivity in vivo to a mutant such as Berg. Just like the binding and stability data in FIGS. 2A, 2B, 3A and 3B, reordering was despite a lack of molecular understanding of the temperature-sensitive phenotype. Regularity is brought to the response surface.

These results suggest the generality of the above reordering approach and its applicability to a wide range of other types of multidimensional data analysis. As will be readily appreciated, many variations and combinations of the features set forth above may be utilized without departing from the invention as set forth in the claims. Such modifications should not be construed as deviations from the spirit and scope of the present invention, but all such modifications are intended to be included within the scope of the patent scope.

[Brief description of drawings]

1 is a flow chart showing a method according to the present invention.

FIG. 2A is a histogram plot of stability data from mutations of the gene V protein at sites I47 and V35 constructed as originally shown.

2B is a rearrangement of the data of FIG. 2A performed by the method according to the invention.

3A is similar to FIG. 2A, but for DNA binding affinity.

3B is a similar rearrangement of the data of FIG. 3A.

FIG. 4A is a histogram plot of phenotypic data from mutations of the gene V protein at sites I47 and V35 constructed as originally shown.

4B is a similar rearrangement of the data of FIG. 4A.

─────────────────────────────────────────────────── ─── Continued front page    (81) Designated countries EP (AT, BE, CH, CY, DE, DK, ES, FI, FR, GB, GR, IE, I T, LU, MC, NL, PT, SE, TR), OA (BF , BJ, CF, CG, CI, CM, GA, GN, GW, ML, MR, NE, SN, TD, TG), AP (GH, G M, KE, LS, MW, MZ, SD, SL, SZ, TZ , UG, ZW), EA (AM, AZ, BY, KG, KZ, MD, RU, TJ, TM), AE, AG, AL, AM, AT, AU, AZ, BA, BB, BG, BR, BY, B Z, CA, CH, CN, CR, CU, CZ, DE, DK , DM, DZ, EE, ES, FI, GB, GD, GE, GH, GM, HR, HU, ID, IL, IN, IS, J P, KE, KG, KP, KR, KZ, LC, LK, LR , LS, LT, LU, LV, MA, MD, MG, MK, MN, MW, MX, MZ, NO, NZ, PL, PT, R O, RU, SD, SE, SG, SI, SK, SL, TJ , TM, TR, TT, TZ, UA, UG, US, UZ, VN, YU, ZA, ZW

Claims (23)

[Claims] 1. A method for selectively varying multivariable molecular synthesis for optimizing functional properties of reaction products, which is obtained by reacting all primary library input variable combinations to the multivariable synthesis. Constructing a primary library of functional property output data and measuring the functional properties for all reaction products, wherein the primary library input variable combinations are all selections for variables adopted one at a time. Order the values for each input variable according to the effect of the input variables on the functional characteristic optimization based on the above primary library of functional characteristic output data, while including the values, while other variable values are kept constant or randomized Obtained by reacting a process with a set of low-frequency sampled secondary library input variable combinations from the ordered input variables Constructing a secondary library of functional characteristic output data and measuring the functional characteristic for all reaction products, wherein the secondary library input variable combination is a low frequency of the ordered input variables. For the variables that are combined two at a time from sampling, while there are other variable values, the other variable values are held constant or randomized, and for optimizing the above functional characteristics. And a step of interpolating between the functional characteristic output data described above. 2. The method of claim 1, further comprising the step of selecting a reaction product having optimal functional properties from the results of said interpolating step. 3. The method of claim 1, wherein at least one input variable among the combinations is randomized when building the primary library. 4. The method of claim 3, wherein all the input variables among the combinations are randomized when building the primary library. 5. The multivariate synthetic reaction is the synthesis of a molecule to be studied for drug action, and the input variables are altered in one or more chemical scaffolds and in the chemical functionalization sites of the scaffolds. The method of claim 1, including a set of parts to be made. 6. The multivariable synthesis reaction is the synthesis of a protein for the study of site-directed mutagenesis, and the input variable comprises the set of amino acid residues to be incorporated at each mutation site. The method described. 7. The method of claim 1, wherein at least one input variable is reaction condition. 8. After the step of ordering the values for each input variable according to the effect of each input variable on the functional characteristic optimization, the method further comprises: data for functional characteristic optimization for at least one input variable. Selecting additional input variable combinations by interpolation of the above, adding the additional input variable combinations to the primary library, and constructing the primary library of the functional characteristic output data before building the secondary library. The method of claim 1, comprising: repeating the steps of constructing and the ordering of the values of each input variable. 9. The method of claim 8, wherein the additional input variable combination comprises a quadratic combination. 10. The method of claim 8, wherein the additional input variable combination comprises a new variable selection. 11. The method of claim 1, wherein the infrequent sampling is ordered infrequent sampling. 12. The method of claim 11, wherein the ordered infrequent sampling is non-uniform. 13. The method of claim 1, wherein the infrequent sampling is random infrequent sampling. 14. The random infrequent sampling is non-uniform.
The method described. 15. The method further comprising: identifying an input variable combination that produces a reaction product having optimal functional characteristics; and reacting the optimal input variable combination to synthesize the reaction product. The method described in 1. 16. After the interpolation step, the method further comprises: (1) (a) infrequently sampling additional secondary input variable combinations from the ordered input variables, and (b) the additional secondary variables. React the next input variable combination, and
(C) expanding the secondary library of functional property output data by measuring the functional properties for all reaction products; and (2) repeating the interpolation process. The method according to item 1. 17. The method of claim 16, wherein the step of expanding the secondary library of functional characteristic output data and the step of repeating the interpolation step are performed iteratively. 18. The method further comprising repeating the step of ordering the values for each input variable according to the effect of the input variables on the functional characteristic optimization based on both the primary and secondary libraries of functional characteristic output data. Item 17. The method according to Item 17. 19. Before iterating the steps of ordering the values for each input variable, the method further repeats the steps of building a primary library to determine an additional primary library of functional characteristic output data. 19. The method of claim 18 including the steps. 20. After the interpolation step, the method further comprises: (1) creating a correlated secondary input variable combination based on the secondary library of functional characteristic output data; and (2) the function. The method of claim 1, comprising repeating the steps of constructing the secondary library of characteristic output data and interpolating between the functional characteristic output data. 21. The method of claim 1, wherein the functional characteristic output data comprises output data for more than one type of functional characteristic, and variable values are independently ordered for each functional characteristic. 22. After the step of interpolating, the method further comprises increasing the number of values selected for at least one variable and repeating the method from the construction of the primary library. The method according to item 1. 23. After the step of interpolating, the method further comprises generating a tertiary library of functional characteristic output data obtained by reacting a set of infrequently sampled tertiary library input variable combinations from the ordered input variables. Constructing and measuring the functional properties for all reaction products, wherein the tertiary library input variable combinations are variables combined three at a time from the infrequent sampling of the ordered input variables. On the other hand, if there is another variable value, the other variable value is held constant or randomized, and a step of interpolating between the functional characteristic output data for optimizing the functional characteristic. The method of claim 1, comprising: