Graphic user interface assisted chemical structure generation

Patent metadata

Field	Value
Publication number	US-2020272702-A1
Application number	US-201916284011-A
Country	US
Kind code	A1
Filing date	Feb 25, 2019
Priority date	Feb 25, 2019
Publication date	Aug 27, 2020
Grant date	—

Abstract

A computer implemented method of generating new chemical compounds is provided. The method includes preparing a feature vector for each of a plurality of chemical compounds for which a chemical or physical property is known. The method further includes compressing each of the feature vectors into a relational vector, and mapping each of the relational vectors to a map having at least two dimensions. The method further includes presenting the map on a display device. The method further includes receiving a selection of a position on the map, wherein the position is converted to a new relational vector, and decompressing the new relational vector to a candidate feature vector. The method further includes generating a new chemical structure from the candidate feature vector.

First claim

1 . A computer implemented method of generating new chemical compounds, comprising:
preparing a feature vector for each of a plurality of chemical compounds for which a chemical or physical property is known; compressing each of the feature vectors into a relational vector; mapping each of the relational vectors to a map having at least 2 dimensions; presenting the map on a display device; receiving a selection of a position on the map, wherein the position is converted to a new relational vector; decompressing the new relational vector to a candidate feature vector; and generating a new chemical structure from the candidate feature vector.

2 . The computer implemented method of claim 1 , wherein each of the feature vectors is compressed into the relational vector by one or more hidden layers of a neural network.

3 . The computer implemented method of claim 2 , wherein the neural network is a stochastic neural network.

4 . The computer implemented method of claim 1 , further comprising synthesizing the new chemical structure.

5 . The computer implemented method of claim 1 , wherein selection of the position on the 2 dimensional map is received from a user, and selection of a chemical and/or physical property is also received from the user.

6 . The computer implemented method of claim 5 , wherein the new relational vector includes coordinate value, σ, representing a dispersion value of a Gaussian distribution from which a compressed feature vector, z, is sampled, and value, μ, representing a mean value of a Gaussian distribution from which the compressed feature vector, z, is sampled.

7 . The computer implemented method of claim 6 , wherein the data-driven substructure feature vector uses SMILES grammar to represent the plurality of chemical compounds, and the predefined component feature vector uses SMILES grammar to represent the predefined chemical substructures.

8 . The computer implemented method of claim 7 , further comprising testing the synthesized candidate structure to determine the actual value for the chemical or physical property.

9 . A system for generating new chemical compounds, comprising:
a display device; memory, wherein a data set of a plurality of chemical compounds for which a chemical or physical property is stored in the memory; a processor device, wherein the processor device is configured to prepare a feature vector for each of a plurality of chemical compounds for which a chemical or physical property is known; a chemical structure generator configured to compress each of the feature vectors into a relational vector; map each of the relational vectors to a map having at least two dimensions; present the map on the display device; receive a selection of a position on the map, wherein the position is converted to a new relational vector; decompress the new relational vector to a candidate feature vector; and generate a new chemical structure from the candidate feature vector.

10 . The system of claim 9 , wherein each of the feature vectors is compressed into the relational vector by one or more hidden layers of a neural network.

11 . The system of claim 10 , wherein the neural network is a stochastic neural network.

12 . The system of claim 11 , wherein selection of the position on the map is received from a user, and selection of a chemical and/or physical property is also received from the user.

13 . The system of claim 12 , wherein the new relational vector includes coordinates value, a, representing a dispersion value of a Gaussian distribution from which a compressed feature vector, z, is sampled, and a value, μ, representing a mean value of a Gaussian distribution from which the compressed feature vector, z, is sampled, and the relational vectors are each displayed as icons having different colors or shading representing differences in the associated value of the selected property on the display device.

14 . The system of claim 13 , wherein the data-driven substructure feature vector uses SMILES grammar to represent the plurality of chemical compounds, and the predefined component feature vector uses SMILES grammar to represent the predefined chemical substructures.

15 . The system of claim 14 , wherein the candidate structure is generated using backbone structuring, atomistic detailing, and bond detailing.

16 . A non-transitory computer readable storage medium comprising a computer readable program for generating new chemical compounds, wherein the computer readable program when executed on a computer causes the computer to perform the steps of:
preparing a feature vector for each of a plurality of chemical compounds for which a chemical or physical property is known; compressing each of the feature vectors into a relational vector; mapping each of the relational vectors to a map having at least two dimensions; presenting the map on a display device; receiving a selection of a position on the map, wherein the position is converted to a new relational vector; decompressing the new relational vector to a candidate feature vector; and generating a new chemical structure from the candidate feature vector.

17 . The computer readable storage medium of claim 16 , wherein each of the feature vectors is compressed into the relational vector by one or more hidden layers of a neural network.

18 . The computer readable storage medium of claim 17 , wherein the neural network is a stochastic neural network.

19 . The computer readable storage medium of claim 16 , wherein selection of the position on the 2 dimensional map is received from a user, and selection of a chemical and/or physical property is also received from the user.

20 . The computer readable storage medium of claim 16 , further comprising instructions for synthesizing the new chemical structure.

Assignees

• IBM

Inventors

• Takeda Seiji

Classifications

• G06N3/047 Primary

  Probabilistic or stochastic networks · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/0475

  Generative networks · CPC title