System and Method for Identifying Similar Molecules

• US 20080004810A1
• Filed: 06/30/2006
• Published: 01/03/2008
• Est. Priority Date: 06/30/2006
• Status: Active Grant

• Current Assignee: Globalfoundries US Incorporated (GlobalFoundries, Inc.)
• Original Assignee: International Business Machines Corporation
• Inventors: Rhodes, James J., Kreulen, Jeffrey Thomas, Kanungo, Tapas, Breyta, Gregory, Boyer, Stephen Kane
• Granted Patent: US 8,140,267 B2
• Time in Patent Office: Days
• US Class Current: 702/19
• CPC Class Codes: G16C 20/40 Searching chemical structures or physicochemical data
• CPC Class Codes: G16C 20/80 Data visualisation

Abstract

A vectorization process is employed in which chemical identifier strings are converted into respective vectors. These vectors may then be searched to identify molecules that are identical or similar to each other. The dimensions of the vector space can be defined by sequences of symbols that make up the chemical identifier strings. The International Chemical Identifier (InChI) string defined by the International Union of Pure and Applied Chemistry (IUPAC) is particularly well suited for these methods.

Claims

1. A method, comprising:

constructing a vector space having dimensions determined by a plurality of chemical identifier strings, wherein the strings are determined by respective chemical compounds; and

constructing a vector for each of the strings, wherein each vector has the dimensions of the constructed vector space.

2. The method of claim 1, further comprising:

computing a measure of similarity between vectors.

3. The method of claim 2, further comprising:

identifying vectors that are identical to each other.

4. The method of claim 2, further comprising:

identifying vectors that are similar to each other; and

identifying similar chemical compounds in view of the similar vectors.

5. The method of claim 2, further comprising ranking vectors according to the computed measure of similarity.

6. The method of claim 1, wherein the strings are InChI strings.

7. The method of claim 1, wherein sparse vector representations are used.

8. At least one tangible computer-useable medium, said at least one medium having a computer-readable program, wherein the program upon being processed on a computer causes the computer to implement the steps of:

constructing a vector space having dimensions determined by a plurality of chemical identifier strings, wherein the strings are determined by respective chemical compounds; and

constructing a vector for each of the strings, wherein each vector has the dimensions of the constructed vector space.

9. A method, comprising:

extracting sequences of symbols from each of a plurality of chemical identifier strings, wherein each string is associated with a chemical; and

defining a vector for each of the strings, the vectors having a common vector space that includes dimensions given by the extracted sequences.

10. The method of claim 9, wherein the sequences are extracted from InChI strings.

11. The method of claim 9, wherein the extracted sequences include consecutive symbols containing carbon connectivity information.

12. The method of claim 11, wherein the extracted sequences that include consecutive symbols include every possible sequence that has no greater than a predetermined number of symbols.

13. The method of claim 9, wherein the extracted sequences include consecutive symbols containing hydrogen connectivity information.

14. The method of claim 9, wherein each of the extracted sequences has no greater than a predetermined number of symbols.

15. The method of claim 9, wherein the vector space includes dimensions defined by information taken from chemical formulae of the chemicals.

16. The method of claim 15, wherein the dimensions of the vector space include dimensions defined by elements of the chemical formulae.

17. The method of claim 9, further comprising searching for chemicals that are at least similar to each other by calculating a similarity value between a given vector and each of a plurality of the defined vectors.

18. The method of claim 9, wherein the vectors are represented by respective sparse vector representations.

19. At least one tangible computer-useable medium, said at least one medium having a computer-readable program, wherein the program upon being processed on a computer causes the computer to implement the steps of:

extracting sequences of symbols from each of a plurality of chemical identifier strings, wherein each string is associated with a chemical; and

defining a vector for each of the strings, the vectors having a common vector space that includes dimensions given by the extracted sequences.

20. A method, comprising:

converting chemical names to respective chemical identifier strings, the strings having a common format;

constructing respective vectors from the strings;

storing at least some of the vectors in at least one memory device; and

searching at least some of the stored vectors to identify certain chemical structures that are at least similar to each other.

21. The method of claim 20, wherein the identified chemical structures include structures that are identical to each other.

22. The method of claim 20, wherein the strings are InChI strings.

23. The method of claim 20, further comprising:

extracting chemical names from the text of different documents; and

identifying particular documents from which said certain chemical structures were extracted.

24. The method of claim 23, wherein said particular documents include patents having named assignees and inventors, the method further comprising identifying at least one of the assignees and the inventors.

25. The method of claim 20, wherein a vector corresponding to a query molecule is used to identify said certain chemical structures.

26. The method of claim 20, comprising evaluating at least one of cosine distance calculations and Tanimoto coefficient calculations to identify said certain chemical structures.

27. The method of claim 20, wherein prior to searching, the vectors are normalized to unit vectors.

28. The method of claim 20, wherein at least one million chemical names are converted to the common string format.

29. The method of claim 20, wherein the chemical names includes IUPAC names, the method further comprising:

converting IUPAC names to respective structures; and

converting said respective structures to respective chemical identifier strings having the common format.

30. The method of claim 20, wherein said respective vectors are constructed by:

extracting sequences of symbols from each of a plurality of chemical identifier strings; and

defining a vector for each of the strings, the vectors having a common vector space that includes dimensions given by the extracted sequences.

31. The method of claim 20, wherein at least some of the chemical names are extracted from at least one document having text that has been tokenized, so that tokens correspond to terms within the document, the method further comprising:

evaluating each token against at least 2 different Markov models to determine respective relative probabilities that the token corresponds to the Markov models, wherein at least one of the Markov models is directed to chemical terms;

for each token, comparing the relative probabilities with each other to determine which Markov model is more likely to be associated with the token; and

identifying tokens most likely to correspond to said at least one Markov model directed to chemical terms, so that chemical terms within the document are identified.

32. The method of claim 20, wherein the stored vectors include sparse vector representations.

33. At least one tangible computer-useable medium, said at least one medium having a computer-readable program, wherein the program upon being processed on a computer causes the computer to implement the steps of:

converting chemical names to respective chemical identifier strings, the strings having a common format;

constructing respective vectors from the strings;

storing at least some of the vectors in at least one memory device; and

searching at least some of the stored vectors to identify certain chemical structures that are at least similar to each other.

34. A method, comprising:

extracting chemical entities from different documents, the chemical entities having different formats with respect to at least one of name and chemical identifier string;

representing the chemical entities as respective chemical identifier strings having a common format;

constructing respective vectors from the commonly formatted chemical identifier strings;

storing at least some of the vectors in at least one memory device; and

searching at least some of the stored vectors.

35. The method of claim 34, said step of constructing respective vectors comprising:

extracting sequences of symbols from each of the commonly formatted chemical identifier strings; and

defining a vector for each of the commonly formatted strings, the vectors having a common vector space that includes dimensions given by the extracted sequences.

36. The method of claim 35, wherein the commonly formatted strings are InChI strings.

37. The method of claim 35, wherein the commonly formatted strings include information in the InChI format, as well as additional information related to functional properties of the chemical entities, the method further including searching the additional information.

38. The method of claim 34, the chemical entities including chemical names, chemical formula, chemical structures, and chemical identifier strings.

39. The method of claim 34, wherein at least some of the chemical entities are extracted from at least one document having text that has been tokenized, so that tokens correspond to terms within the document, the method further comprising:

evaluating each token against at least 2 different Markov models to determine respective relative probabilities that the token corresponds to the Markov models, wherein at least one of the Markov models is directed to chemical terms;

for each token, comparing the relative probabilities with each other to determine which Markov model is more likely to be associated with the token; and

identifying tokens most likely to correspond to said at least one Markov model directed to chemical terms, so that chemical terms within the document are identified.

40. At least one tangible computer-useable medium, said at least one medium having a computer-readable program, wherein the program upon being processed on a computer causes the computer to implement the steps of:

extracting chemical entities from different documents, the chemical entities having different formats with respect to at least one of name and chemical identifier string;

representing the chemical entities as respective chemical identifier strings having a common format;

constructing respective vectors from the commonly formatted chemical identifier strings;

storing at least some of the vectors in at least one memory device; and

searching at least some of the stored vectors.

41. A system, comprising:

said at least one medium of claim 40; and

a processor in communication with said at least one medium, the processor and said at least one medium together forming at least part of a computer system.

42. A computer-implemented method, comprising:

processing the program of said at least one medium of claim 40 to implement the steps of claim 40; and

delivering to a client output resulting from implementing the steps of claim 40.