Method, system, and computer program product for analyzing combinatorial libraries

• US 7,054,757 B2
• Filed: 01/29/2002
• Issued: 05/30/2006
• Est. Priority Date: 01/29/2001
• Status: Expired due to Fees

• Current Assignee: Johnson & Johnson Pharmaceutical Research & Development LLC (Johnson & Johnson)
• Original Assignee: Johnson & Johnson Pharmaceutical Research & Development LLC (Johnson & Johnson)
• Inventors: Agrafiotis, Dimitris K, Salemme, Francis R, Lobanov, Victor S
• Primary Examiner(s): Zeman, Mary K.
• Application Number:
  US10/058,216
• Publication Number:
  US 20020143476A1
• Time in Patent Office: 1,582 Days
• Field of Search: 702/19, 702/22
• US Class Current: 702/19
• CPC Class Codes: B01J 2219/00686 Automatic
• CPC Class Codes: B01J 2219/00689 using computers
• CPC Class Codes: B01J 2219/00695 Synthesis control routines, e.g. using computer programs
• CPC Class Codes: B01J 2219/007 Simulation or vitual synthesis
• CPC Class Codes: B01J 2219/00702 Processes involving means for analysing and characterising the products
• CPC Class Codes: B01J 2219/0072 Organic compounds
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/02 Neural networks
• CPC Class Codes: G16B 35/00 ICT specially adapted for in silico combinatorial libraries of nucleic acids, proteins or peptides
• CPC Class Codes: G16C 20/60 In silico combinatorial chemistry
• CPC Class Codes: G16C 20/62 Design of libraries
• CPC Class Codes: G16C 20/70 Machine learning, data mining or chemometrics
• CPC Class Codes: Y10S 707/99943 Generating database or data structure, e.g. via user interface

Abstract

The invention provides for in silico analysis of a virtual combinatorial library. Mapping coordinates for a training subset of products in the combinatorial library, and features of their building blocks, are obtained. A supervised machine learning approach is used to infer a mapping function ƒ that transforms the building block features for each product in the training subset of products to the corresponding mapping coordinates for each product in the training subset of products. The mapping function ƒ is then encoded in a computer readable medium. The mapping function ƒ can be retrieved and used to generate mapping coordinates for any product in the combinatorial library from the building block features associated with the product.

Claims

1. A method for generating coordinates for products in a combinatorial library based on features of corresponding building blocks, the method comprising the steps of:

(1) obtaining mapping coordinates for a subset of products in the combinatorial library, wherein distances between the mapping coordinates represent relationships between the products;

(2) obtaining building block features for the subset of products in the combinatorial library;

(3) using a supervised machine learning approach to infer a mapping function ƒ

that transforms the building block features for each product in the subset of products to a corresponding mapping coordinate for each building block feature of each product in the subset of products; and

(4) encoding the mapping function ƒ

in a computer readable medium, whereby the mapping function ƒ

is used to generate coordinates corresponding to additional products in the combinatorial library from building block features associated with the additional products.

2. The method according to claim 1, further comprising the step of:

(5) providing building block features for at least one additional product to the mapping function ƒ

wherein the mapping function ƒ

outputs generated mapping coordinates for the additional product.

3. The method according to claim 1, wherein step (1) comprises generating the mapping coordinates for the subset of products.

4. The method according to claim 3, wherein step (1) further comprises the steps of:

(a) generating an initial set of mapping coordinates for the subset of products;

(b) selecting two products from the subset of products;

(c) determining a relationship between the two products;

(d) refining the initial set of mapping coordinates of one product selected in step (1)(b) based on the relationship and the corresponding distance between the products on the nonlinear map; and

(e) repeating steps (1)(b) and (1)(c) for additional products until a stop criterion is obtained.

5. The method according to claim 1, wherein step (1) comprises calculating the mapping coordinates for the subset of products using a dimensionality reduction algorithm.

6. The method according to claim 1, wherein step (1) comprises retrieving the mapping coordinates for the subset of products from a computer readable medium.

7. The method according to claim 1, wherein step (2) comprises the step of:

using a laboratory measured value as a feature for each building block in at least one variation site in the combinatorial library.

8. The method according to claim 1, wherein step (2) comprises the step of:

using a computed value as a feature for each building block in at least one variation site in the combinatorial library.

9. The method according to claim 1, wherein at least some of the building block features represent reagents used to construct the combinatorial library.

10. The method according to claim 1, wherein at least some of the building block features represent chemical fragments of reagents used to construct the combinatorial library.

11. The method according to claim 1, wherein at least some of the building block features represent modified chemical fragments of reagents used to construct the combinatorial library.

12. The method according to claim 1, wherein the mapping function ƒ

is encoded as a neural network.

13. The method according to claim 1, wherein the mapping function ƒ

is a set of mapping functions ƒ

₁ through ƒ

_n, each encoded as a neural network.

14. The method of claim 1, wherein step (3) comprises:

(a) placing the selected training subset of products on an m-dimensional nonlinear map using randomly assigned coordinates;

(b) selecting a pair of the products having a similarity relationship;

(c) revising the coordinates of at least one of the selected pair of products based on the similarity relationship and the corresponding distance between the products on the nonlinear map; and

(d) repeating steps (b) and (c) for additional pairs of the products until the distances between the products on the m dimensional nonlinear map are representative of the similarity relationships between the products.

15. A computer program product for generating coordinates for products in a combinatorial library based on features of corresponding building blocks, said computer program product comprising a computer useable medium having computer program logic recorded thereon for controlling a processor, said computer program logic comprising:

a procedure that enables said processor to obtain mapping coordinates for a subset of products in the combinatorial library, wherein distances between the mapping coordinates represent similarity/dissimilarity of the products;

a procedure that enables said processor to obtain building block features for the subset of products in the combinatorial library;

a procedure that enables said processor to use a supervised machine learning approach to infer a mapping function ƒ

that transforms the building block features for each product in the subset of products to a corresponding mapping coordinate for each building block feature of each product in the subset of products; and

a procedure that enables said processor to encode the mapping function ƒ

in a computer readable medium, whereby the mapping function ƒ

is used to generate coordinates corresponding to additional products in the combinatorial library from building block features associated with the additional products.

16. The computer program product of claim 15, further comprising:

a procedure that enable said processor to provide building blocks features for at least one additional product to the mapping function ƒ

wherein the mapping function ƒ

outputs generated mapping coordinates for the additional product.

17. The computer program product of claim 15, wherein said procedure that enables said processor to obtain mapping coordinates comprises:

a procedure that enables said processor to generate an initial set of mapping coordinates for the subset of products;

a procedure that enables said processor to select two products from the subset of products;

a procedure that enables said processor to determine a relationship between the two products;

a procedure that enable said processor to refine the initial set of mapping coordinates of at least one product selected based on the relationship and the corresponding distance between the products on the nonlinear map; and

a procedure that enables said processor to continue selecting two products at a time and refining the mapping coordinates of at least one product selected until a stop criterion is obtained.

18. The computer program product of claim 15, wherein a laboratory measure value is used as a feature for each building block in at least one variation site in the combinatorial library.

19. The computer program product of claim 15, wherein a computed value is used as a feature for each building block in at least one variation site in the combinatorial library.

20. The computer program product of claim 15, wherein at least some of the building block features represent reagents used to construct the combinatorial library.

21. The computer program product of claim 15, wherein at least some of the building block features represent chemical fragments of reagents used to construct the combinatorial library.

22. The computer program product of claim 15, wherein at least some of the building block features represent modified chemical fragments of reagents used to construct the combinatorial library.

23. The computer program product of claim 15, wherein the mapping function ƒ

is encoded as a neural network.

24. The computer program product of claim 15, wherein the mapping function ƒ

is a set of mapping functions ƒ

₁ through ƒ

_n, each encoded as a neural network.

25. A method for generating coordinates for products in a combinatorial library based on features of corresponding building blocks, the method comprising the steps of:

(1) obtaining mapping coordinates for a subset of products in the combinatorial library, wherein distances between the mapping coordinates represent similarity/dissimilarity relationships between the products;

(2) obtaining building block features for the subset of products in the combinatorial library;

(3) using a supervised machine learning approach to infer a mapping function ƒ

that transforms the building block features for each product in the subset of products to a corresponding mapping coordinate for each building block feature of each product in the subset of products; and

(4) encoding the mapping function ƒ

in a computer readable medium, whereby the mapping function ƒ

is used to generate coordinates corresponding to additional products in the combinatorial library from building block features associated with the additional products.