JP4895689B2 - System and method for retrieving chemical structure from large-scale chemical structure database at high speed - Google Patents

Description

  The present invention relates to a system and method for retrieving a chemical structure at high speed from a large-scale chemical structure database.

  The chemical structure database is not only an indispensable tool in modern chemical and drug discovery research, but also an indispensable tool for patent information and reagent management. For example, in the chemical structure search for Chemical Abstracts System, SciFinder [http: // www. jaici. or. jp / sci / SCHOLAR / index. html] is the ISIS [http: // www. mdli. com /] etc. are widely used.

Since 1960, many scientists have conducted thorough research on the development of chemical structure search algorithms. The partial structure search is an operation for determining whether or not a specified query structure is included in a given target structure. Using the terminology of graph theory, substructure searches query graph (G _Q) is a work to determine whether a subgraph isomorphic target graph (G _T), between the subgraph of G _Q and G _T It is known that searching for the isomorphism of is an NP complete problem [Non-Patent Document 1]. Therefore, it is generally a very difficult task to quickly determine isomorphism, but many substructure search techniques solve this problem by using the backtracking method efficiently [non- Patent Document 2].

However, in order to perform a chemical structure search from a chemical structure database of 50,000 or more, the backtracking method alone requires a lot of search time (several tens of seconds) and is not practical. Therefore, a technique called “screening” has been developed that quickly removes graphs (chemical structures) that are clearly not isomorphic before performing the backtracking method.
Usually, the screening method represents a chemical structure with a bit character string. Each bit means an arbitrary fragment (benzene ring, amide group, etc.), where 1 indicates that the fragment is present in the chemical structure and 0 indicates that it is not present. The target structure in the chemical structure database is generated in advance and this bit character string is stored in the database, and the query structure bit string is generated at the time of retrieval. Next, the bit string of the query structure and the bit string of the target structure are sequentially compared, and at least one 1-bit bit present in the bit string of the query structure is present in the bit string of the target structure. Otherwise, it is removed as no possible isomorphism. Since this calculation can be performed using bit operators (AND, OR, XOR), it can be processed at a very high speed [Non-Patent Document 3]. In general, by using a screening method using a bit character string and an improved version thereof, a speed increase of 10 to 20 times is realized as a whole, so that it is currently used in many structure search systems. .

  Usually, speeding up of the search for a database is realized by using an index (corresponding to a book index). Since the index is intended for numerical values and character strings, it cannot be used effectively in existing chemical structure searches that require operations between bit character strings. Therefore, in many chemical structure searches, it is necessary to repeat bit operations sequentially from the beginning to the end of the database, which is the biggest bottleneck for speeding up.

On the other hand, the most widely used search systems for large-scale databases in recent years are Google and Yahoo! This is a full-text search system that “searches a document including a specific character string from a large number of documents” represented by a search engine. The speed of the search speed of the full-text search system is self-evident at a level that anyone using a search engine can agree with. Normally, in a full-text search system, a document is decomposed into words using morphological analysis [Non-patent Document 4] or N-gram method [Non-Patent Document 5], and then this word is indexed by a transposed index method or the like. . By this work, which word is present in which document can be searched at high speed.
M.M. Garey, D.D. Johnson, Computers and Intractability; A Guide to the Theory of NP-Completeness: H. Freeman, New York, 1979. J. et al. Xu, J. et al. Chem. Inf. Comput. Sci. 1996, 36, 25-34. M.M. F. Lynch, Chemical Information Systems, J. MoI. E. Ash, E.M. Hyde eds. Elis Horwood, Chichester, 1985, 88-93. Toru Hisamitsu and Yoshihiko Nitta, “Efficient Verb Utilization Processing in Japanese Morphological Analysis” Report of Information Processing Society of Japan, 94-NL-103, September 1994, 1-7. Kendo Odo, Katsunobu Ito, Kiyohiro Shikano, Satoshi Nakamura, “Study on parameter reduction based on entropy of N-gram model” IPSJ Journal 2001, Vol. 42 no. 2, 327-333

In chemical structure search, simultaneous access requests via the Web by multiple users and the number of compounds registered in the database tend to increase (enlarge) year by year, and faster chemical structure search methods than ever before are required. ing. Furthermore, the development of search methods that can achieve high speeds without relying on expensive commercial databases and without using special hardware such as supercomputers is the benefit of the so-called “cheap revolution”. (With the rapid increase in computer performance and price, and the emergence of high-performance free software provided as open source, it has become possible to develop high-performance applications at low cost. Details of the “Cheap Revolution” Can be directly received by Nobuo Umeda's “Web Evolution” Chikuma New Book.), Enabling the development of an overwhelmingly high-cost chemical structure search system.
From the above, the development of a fast and inexpensive chemical structure search method not only satisfies the simultaneous access request of multiple users to a large-scale database, but also a web application system using chemical structure search that has not been developed yet. It is expected to be available.

In view of the above problems, as a result of intensive studies, if the chemical structure can be expressed as “document” and the partial structure can be expressed as “word”, the existing full-text search system can be used as it is, and Google and Yahoo can be used as they are. ! We have found that a high-level search system can be realized even in chemical structure search, and have developed the ITMChemString (ITCS) method that enables “screening” in chemical structure search to be realized by full-text search.
The gist of the present invention is as follows.
[1] The chemical structure of a compound input to a computer is expressed as a tree structure composed of nodes corresponding to atoms and edges corresponding to bonds between atoms, and a root node is selected from one of the nodes, and the root Means for performing path determination from a node by depth-first search, and characterizing the chemical structure of the compound according to the determined path;
Based on the tree structure, means for characterizing all partial structures constituting the compound;
A storage medium for recording and storing a character string representation of the chemical structure of the obtained compound and a character string representation of the partial structure of the compound together with a unique ID for identifying the compound as a chemical structure database A chemical structure search system for searching for compounds having a partial structure of
[2] 1) The chemical structure of the compound input to the computer is expressed as a tree structure composed of nodes corresponding to atoms and edges corresponding to bonds between atoms, and a root node is selected from one of the nodes. A path is determined by depth-first search from the root node, and the chemical structure of the compound is converted into a character string according to the determined path; and 2) all the constituents of the compound are configured based on the tree structure. Sending a search request to the chemical structure search system according to the above [1] using a character string representation of the partial structure of the compound obtained by characterizing the partial structure as a query;
Using the chemical structure database included in the chemical structure search system, a full text search process is performed on a search request thrown by a computer, and a unique ID for identifying a compound as a search result and the character of the chemical structure of the corresponding target compound Returning a columnized representation;
A compound having a predetermined partial structure, including a step of performing a chemical structure search process on the obtained character string representation of the ID and the chemical structure of the target compound and presenting an ID of the compound having the partial structure as a search result How to search.

  In the chemical structure search of the present invention, the chemical structure is expressed as “document” and the partial structure is expressed as “word”. Therefore, the existing full-text search system can be used as it is, and preferably Google or Yahoo. ! A high-speed search system can also be realized in chemical structure search.

  In the present specification, chemical structure search means partial structure search and complete structure search of a compound.

  The present invention provides a chemical structure retrieval system for retrieving a compound having a predetermined partial structure as described above. Details of the processing performed by each of the above means are as described later.

  The means for converting each character string may be manual or information processing means such as a computer. However, considering the efficiency of processing, information processing means such as a computer is used, It is preferable to construct a program or system in which the chemical structure is input and the subsequent processing is automatically performed by a computer.

  As a storage medium for recording and storing a chemical structure database, a character string representation of a chemical structure of a compound and a character string representation of a partial structure of a compound are recorded as a database together with a unique ID for identifying the compound. Any storage medium can be used as long as it can be stored. For example, as such a storage medium, a hard disk, a nonvolatile memory, a magnetic disk, an optical disk, a magnetic tape and the like arranged inside and outside the computer can be cited.

  In the chemical structure database, in addition to the unique ID for identifying the compound, the character string expression of the chemical structure of the compound, and the character string expression of the partial structure of the compound, other information related to the compound It may be recorded and stored in association with a unique ID for identifying. Such information includes specific physical properties of compounds such as melting point, boiling point, molecular weight, log P, molecular surface area, reactivity, structure and reaction path information.

  An outline of the ITCS method of the present invention is shown in FIG.

In the ITCS method, an ITCS is first generated from a target compound by an “ITCS generation process” (details will be described later). ITCS is a character string (corresponding to a document in a normal full-text search) indicating a chemical structure of a compound.
Next, an ITCS WORD is generated by an “ITCS WORD generation process” (details will be described later). ITCS WORD enumerates all partial structures constituting a chemical structure and converts them into character strings (corresponding to words in ordinary full-text search).
The generated ITCS and ITCS WORD are stored in the database together with the ID and other additional information. The number of target compounds stored here is 1 to tens of millions of compounds, and relational databases such as PostgreSQL, MySQL, Oracle, etc. can be used.

Query compounds can be obtained from general chemical structure input means (for example, chemical structure drawing software (for example, ChemDraw, ISIS / Draw, etc.) used in the field running on a general-purpose computer from general-purpose molecules such as sdf format and mol2 format. ITCS and ITCS WORD are generated in the same manner as the target compound.
Next, a query request is sent to the database using the query compound ITCS WORD as a query. In the database, a search request is sent to the “full text search process” for data corresponding to the search request (ITCS WORD) thrown, and a response (ID) is obtained. The “full-text search process” here is a method known to those skilled in the art, and the method is not particularly limited, and a system that is normally used for full-text search can be used as appropriate. For example, if it is used outside the database, Namazu, Rast, Estraer, etc. can be used, and if it is used inside the database, Tsearch2 etc. can be used in PostgreSQL.
The database returns the ID obtained by the “full-text search process” and the corresponding ITCS as a response to the query.

  The process so far is “screening”.

Finally, ID + ITCS obtained by screening and ITCS of the query compound are subjected to a structure search using a “chemical structure search process”, and hit compound IDs and various information (for example, melting point, boiling point, molecular weight, logP, molecule Indicate physical properties such as surface area, specific physical properties, reactivity, structure and reaction path information.
The “chemical structure search process” here can use a known backtracking method. The “chemical structure search process” can be used not only outside the database but also incorporated inside the database.

  Through the above process, it is possible to retrieve a chemical structure from a large-scale database at high speed.

"ITCS generation process"
As the character string (linear) notation of a chemical structure, WLN, SMILES, ROSDAL, and the like are known and widely used. In the present invention, instead of using a known method from the viewpoint of extensibility, an ITCS notation has been developed independently. The ITCS generation process is shown in FIG.

Step 1:
The chemical structure is prepared in a general-purpose molecular structure interchange format such as sdf format or mol2 format. Here, the bond having the property as an aromatic is identified using the Hückel rule.

Step 2:
Treat the chemical structure as a tree structure, with nodes corresponding to atoms and edges corresponding to bonds. The root node can start with any atom, the node holds the atomic species, the atom ID number, the visit number, and the edge holds the bond order. Here, the atom ID number may be assigned to each atom in an arbitrary way when creating a chemical structure, but is usually assigned depending on the software used when preparing the molecular structure interchange format. It is.

Step 3:
A route is determined from the root node of the tree created in step 2 by a depth-first search. “Depth-first search” is a known algorithm. At this time, the nodes are numbered from 0 according to the order of the nodes to be visited (visit number) and are held in the nodes.

Step 4:
An ITCS is created according to the route created in step 3.
ITCS rules:
・ Atoms are expressed by atomic names (character strings such as C, N, and O).
The bond is represented by a double bond “d”, a triple bond “t”, an aromatic bond “a”, and a single bond “” (nothing).
・ Atoms that have already been visited on the route are represented by a visit number, not an atom name. At this time, if the node at the beginning of the next visited edge is different from the node at the end of the currently visited edge, the atom at the beginning of the next edge is represented by “<visit number>”. Furthermore, if the atom at the end of the next edge has already been visited, this atom is represented by '[visit number]'. For example, referring to step 3 in FIG. 2, if the currently visited edge is N (1) -C (2), the next visited edge is N (1) -C (3). In this case, the last edge currently visited is C (2), and the next visited edge is N (1). Therefore, it is <1> because it is not the same node and N (1) has already been visited.
The case of using [] will be described below using cyclohexane as an example.

If the currently visited edge is C (4) -C (5) and the next visited edge is C (0) -C (5), the start of the next edge and the end of the current edge are different and Since C (0) has already visited, it becomes <0>. Furthermore, since the node C (5) at the end of the next edge has already been visited, the next edge is expressed as <0> [5].
Finally, add '/' (slash) and add atomic ID numbers separated by ',' (comma) according to the order of visit numbers. Normally, when a chemical structure is converted into a character string, the atom ID number on the chemical structure file that is the conversion source cannot be retained after the conversion into a character string. Therefore, in the present invention, the atomic ID number can be retained by adding it after the slash. This makes it possible to match the atoms in the chemical structure search using ITCS with the atoms in the conversion source chemical structure file, and to perform highlighting of the matching atoms on the conversion source chemical structure file. it can.

"ITCS WORD generation process"
The purpose of this process is to enumerate all partial structures constituting an arbitrary chemical structure and convert them into character strings (corresponding to “words” in a normal full-text search). The main parts of the ITCS WORD generation process are shown in FIGS.

Step 1:
A tree created by the ITCS generation process is referred to as a “basic tree”.

Step 2:
Based on the basic tree, a “growth tree” is constructed based on the following “growth tree” construction rules.
Continue until the tree no longer grows based on the “Growth Tree” construction rules. However, the depth of the tree needs to be set in advance (set as n_depth) and can be set from 1 to the number of bonds constituting the chemical structure. Considering the performance of the current computer resources, it is usually set as 4-7.
The “growing tree” construction rule / base node is selected based on the depth-first search order for the growing tree, and the initial base node is the root node on the “basic tree”.
The addition of a node and an edge to the base node means that the node on the “basic tree” corresponding to the base node and its child node and the edge to which they belong, and if an ancestor node exists, the ancestor node and its child node and This is done by copying the edges to which they belong to the base node. However, if the ancestor node, its child node, and the edge to which they belong already exist on the path from the root node to the base node on the growth tree, they are not added. Here, the ancestor node indicates a node located on the route from the base node to the root node.

Step 3:
List all paths from the root node to the end node of the growth tree constructed in step 2. Then, by cutting each route from the root node to a depth of 1, 2,..., N_depth−1, a route corresponding to the partial structure is generated.
All the paths generated here are converted into character strings using the same algorithm as the “ITCS generation process”. The character string generated here is called ITCS WORD.

Step 4:
The processing of Step 1 to Step 3 is repeatedly performed using all atoms constituting the compound structure as root nodes. By this processing, all partial structures configured up to the n_depth length in the chemical structure can be converted into character strings as ITCS WORD.

Step 5:
Since one chemical structure can be expressed by a plurality of ITCS WORDs, those ITCS WORDs are arranged lexicographically in alphabetical order and the largest one is used as a representative. For example, although CCN and NCC show the same structure, if they are arranged in lexicographical alphabetical order, NCC> CCN, and its representative is NCC. Thus, one structure corresponds to one ITCS WORD.

Step 6:
When a plurality of the same partial structures (ITCS WORD) exist in one chemical structure, a numerical value is added after the ITCS WORD. However, the maximum value is 6 and 1 is omitted. For example, ITCS WORD generated when there are three SCO <1> N is SCO <1> N, SCO <1> N2, and SCO <1> N3. Furthermore, a special partial structure that cannot be expressed by the set n_depth length can also be added as an external ITCS WORD at this step. For example, continuous carbon connections such as CCCCCCCCCCCCCC and CCNCCCNCCNCNCNCNCCN, and continuous repeated structures.
However, it is sufficient that the ITCS WORD used as a query is only an ITCS WORD having an n_depth length and an external ITCS WORD (since the partial structure below that is included in the ITCS WORD having an n_depth length).
When n_depth is set as the number of bonds constituting the chemical structure, in principle, all partial structures can be converted into character strings as ITCS WORDs, so that the chemical structure search is completed only by screening. Therefore, it is not necessary to use an existing chemical structure search such as a backtrack method. However, as n_depth increases, ITCS WORD increases exponentially. Therefore, considering the performance of current computer resources, 4-7 is appropriate at present. However, if the computer performance improves, n_depth can be increased.

Example of chemical structure search The following query structure was used as an example of chemical structure search.

The ITCS and ITCS WORD of the query structure are shown below.
ITCS:
CNCNCCCC <4> C <1> [6] <0> CaCaCC <10> aCaCaC <8> a [14] <0> dO / 1,2,7,10,4,9,8,13,3,6 , 11, 16, 15, 14, 12, 5
ITCS WORD:
“CaCaCaCaCaCa [0] NCCaCaC NCCaC <2> aC NCCaC <0> C NCC <0> CC NCC <0> C <0> C NCCNC CaCaCC <0> C CaCaCaCaC CaCaCaC <2> C CaCaCaCC OdCNCC OdCNC <2> COD <1> C OdCN <1> CaC OdCCaCaC OdCCaC <2> aC NCCaCaC2 NCCaC <0> C4 NCC <0> CC4 NCC <0> C <0> C5 NCCNC6 CaCaCaCaC6 CaCaCaC <2> C4 CaCaCaCC2 OdCCC2 OdCNCC <<1> CaC2 OdCCaCaC2 "

  The effectiveness of the ITCS method was verified by examining the difference in the number of data of compounds contained in the database and the search time when the ITCS method was used and when it was not used. Here, the case where the ITCS method is not used indicates a search only by chemical structure search by the backtrack method. The results are shown in the following table and FIG. Examples of hit compounds are also shown below.

  From these results, when the ITCS method was used, the search speed was improved 100 times or more (when the number of stored data was 100,000) compared with the backtrack method alone. Therefore, according to the present invention, it is possible to search a compound having a predetermined partial structure from a large-scale chemical structure database at high speed.

It is a schematic diagram of the system based on the chemical structure search method of the present invention. It is a figure explaining an ITCS production | generation process. It is a figure explaining an ITCS WORD production | generation process. It is a continuation of FIG. It is a graph which shows the result of having investigated how much difference arises in the data number of compounds contained in a database, and search time, when not using ITCS method.

Claims (3)

The chemical structure of the compound input to the computer is expressed as a tree structure consisting of nodes corresponding to atoms and edges corresponding to the bonds between atoms. A root node is selected from one of the nodes, and a depth from the root node is selected. Means for performing route determination by priority search, and characterizing the chemical structure of the compound according to the determined route;
Based on the tree structure, up to a length of n_depth (where n_depth is the number of bonds constituting the chemical structure), which is a depth of a preset tree constituting the compound It means for stringifying all parts structure,
A storage medium for recording and storing a character string representation of the chemical structure of the obtained compound and a character string representation of the partial structure of the compound together with a unique ID for identifying the compound as a chemical structure database A chemical structure search system for searching for compounds having a partial structure of 1) The chemical structure of a compound input to a computer is expressed as a tree structure composed of nodes corresponding to atoms and edges corresponding to bonds between atoms, a root node is selected from one of the nodes, and the root node Determining a route by depth-first search from the character string, and characterizing the chemical structure of the compound according to the determined route; and
2) Characterizing all partial structures having a length of n_depth set for the chemical structure search system according to claim 1 that constitute the compound based on the tree structure;
Throwing a search request to the chemical structure search system according to claim 1, using as a query the character string representation of the partial structure of the compound obtained by
Using the chemical structure database included in the chemical structure search system, a full text search process is performed on a search request thrown by a computer, and a unique ID for identifying a compound as a search result and the character of the chemical structure of the corresponding target compound Returning a columnized representation;
A compound having a predetermined partial structure, including a step of performing a chemical structure search process on the obtained character string representation of the ID and the chemical structure of the target compound and presenting an ID of the compound having the partial structure as a search result How to search. 3. The method according to claim 2, wherein in the step of sending the search request, a character string representation of a special partial structure that constitutes the input compound and cannot be expressed by the length of the n_depth is also used as the query. .

Images