# US 2022/0285038 A1 · Prediction method, electronic device, and storage medium

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2022/0285038 A1
- Document ID: US20220285038A1
- Status: Published application
- Kind: A1
- Application number: 17/739,541
- Filed: 2022-05-09
- Published: 2022-09-08
- Earliest claimed: 2019-11-18
- Assignees: Beijing Sensetime Technology Development Co., Ltd.
- Applicants: Beijing Sensetime Technology Development Co., Ltd.
- Inventors: Qiao LIU; Zhiqiang HU
- Web page: https://patentgrep.com/patent/US20220285038A1

## CPC

- G16H 10/40: for data related to laboratory analysis, e.g. patient specimen analysis
- G16H 70/40: relating to drugs, e.g. their side effects or intended usage
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G16B 25/10: Gene or protein expression profiling; Expression-ratio estimation or normalisation
- G16C 20/30: Prediction of properties of chemical compounds, compositions or mixtures
- G16H 20/10: relating to drugs or medications, e.g. for ensuring correct administration to patients
- G16H 40/63: for local operation
- G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert systems
- G16H 50/70: for mining of medical data, e.g. analysing previous cases of other patients
- G16B 20/00: ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
- G16C 20/70: Machine learning, data mining or chemometrics
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G16B: BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16C: COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE

## Abstract

A prediction method, an electronic device and a storage medium are provided. The method includes that: substance features of a substance to be tested are determined according to a molecular structure of the substance to be tested; feature extraction is performed on a diseased cell of a target category to obtain at least one cell feature of the diseased cell; and a response result of the substance to be tested against the diseased cell is predicted according to the substance features and the at least one cell feature.

## Claims

### Claim 1

A prediction method, comprising:
according to a molecular structure of a substance to be tested, determining substance features of the substance to be tested;
performing feature extraction on a diseased cell of a target category to obtain at least one cell feature of the diseased cell; and
according to the substance features and the at least one cell feature, predicting a response result of the substance to be tested against the diseased cell.

### Claim 2 (depends on 1)

The prediction method of claim 1, wherein determining the substance features of the substance to be tested according to the molecular structure of the substance to be tested comprises:
according to the molecular structure of the substance to be tested, constructing a structure feature map of the substance to be tested, wherein the structure feature map includes at least two nodes and lines between the nodes, each node represents an atom in the molecular structure, and each line represents an atomic bond in the molecular structure; and
according to the structure feature map, determining the substance features of the substance to be tested.

### Claim 3 (depends on 2)

The prediction method of claim 2, wherein determining the substance features of the substance to be tested according to the structure feature map comprises:
according to the structure feature map, obtaining a first adjacent matrix and a first feature matrix of the substance to be tested, wherein the first adjacent matrix represents neighbor relationships between atoms of the substance to be tested, and the first feature matrix represents attribute data of each atom of the substance to be tested; and
according to the first adjacent matrix and the first feature matrix, obtaining the substance features of the substance to be tested.

### Claim 4 (depends on 3)

The prediction method of claim 3, wherein obtaining the substance features of the substance to be tested according to the first adjacent matrix and the first feature matrix comprises:
constructing a complementary matrix of the first adjacent matrix according to a preset input dimension and a dimension of the first adjacent matrix, and constructing a complementary matrix of the first feature matrix according to the preset input dimension and a dimension of the first feature matrix;
concatenating the first adjacent matrix and the complementary matrix of the first adjacent matrix to obtain a second adjacent matrix with the preset input dimension, and concatenating the first feature matrix and the complementary matrix of the first feature matrix to obtain a second feature matrix with the preset input dimension; and
performing graph convolution processing on the second adjacent matrix and the second feature matrix to obtain the substance features of the substance to be tested.

### Claim 5 (depends on 4)

The prediction method of claim 4, wherein, in the second adjacent matrix, the first adjacent matrix has no adjacent relationship with the complementary matrix of the first adjacent matrix.

### Claim 6 (depends on 4)

The prediction method of claim 4, wherein concatenating the first adjacent matrix and the complementary matrix of the first adjacent matrix to obtain the second adjacent matrix with the preset input dimension, and concatenating the first feature matrix and the complementary matrix of the first feature matrix to obtain the second feature matrix with the preset input dimension comprises:
according to the first adjacent matrix and the complementary matrix of the first adjacent matrix, constructing a first conjunction matrix, wherein elements in the first conjunction matrix are all preset values;
connecting the first adjacent matrix and the complementary matrix of the first adjacent matrix through the first conjunction matrix to obtain the second adjacent matrix with the preset input dimension; and
connecting the first feature matrix and the complementary matrix of the first feature matrix to obtain the second feature matrix with the preset input dimension.

### Claim 7 (depends on 1)

The prediction method of claim 1, wherein performing feature extraction on the diseased cell of the target category to obtain the at least one cell feature of the diseased cell comprises at least one of:
performing feature extraction on genomic mutation of the diseased cell to obtain a genome feature of the diseased cell;
performing feature extraction on gene expression of the diseased cell to obtain a transcriptome feature of the diseased cell; or
performing feature extraction on Deoxyribonucleic Acid (DNA) methylation data of the diseased cell to obtain an epigenome feature of the diseased cell.

### Claim 8 (depends on 1)

The prediction method of claim 1, wherein predicting the response result of the substance to be tested against the diseased cell according to the substance features and the at least one cell feature comprises:
concatenating the substance features and the at least one cell feature to obtain a combined feature after concatenation; and
performing convolution processing on the combined feature to obtain a predicted response result of the substance to be tested against the diseased cell.

### Claim 9 (depends on 8)

The prediction method of claim 8, wherein the at least one cell feature includes at least one genome feature, at least one transcriptome feature, and at least one epigenome feature, and
wherein concatenating the substance features and the at least one cell feature to obtain the combined feature after concatenation comprises:
concatenating the substance features and at least one of the genome feature, the transcriptome feature or the epigenome feature to obtain the combined feature after concatenation.

### Claim 10 (depends on 1)

The prediction method of claim 1, wherein the method is implemented by a neural network and comprises:
training the neural network based on a preset training set, wherein the preset training set comprises a plurality of groups of sample data, and each group of sample data comprises: a structure feature map of a sample substance, genomic mutation of a sample diseased cell, gene expression of the sample diseased cell, Deoxyribonucleic Acid (DNA) methylation data of the sample diseased cell, and a labeled response result of the sample substance against the sample diseased cell.

### Claim 11 (depends on 10)

The prediction method of claim 10, wherein the neural network comprises a first feature extraction network, a second feature extraction network and a prediction network; and
wherein training the neural network based on the preset training set comprises:
performing feature extraction on the structure feature map of the sample substance through the first feature extraction network to obtain sample substance features of the sample substance;
extracting, through the second feature extraction network, at least one sample genome feature corresponding to the genomic mutation of the sample diseased cell, at least one sample transcriptome feature corresponding to the gene expression of the sample diseased cell, and at least one sample epigenome feature corresponding to the DNA methylation data of the sample diseased cell;
performing, through the prediction network, convolution processing on a sample combined feature obtained after concatenation of the sample substance features, the sample genome feature, the sample transcriptome feature and the sample epigenome feature, to obtain a response result of the sample substance against the sample diseased cell;
determining a predicted loss of the neural network according to the response result and the labeled response result; and
training the neural network according to the predicted loss.

### Claim 12

An electronic device, comprising:
a processor; and
a memory, configured to store instructions that, when executed by the processor, cause the processor to perform the following operations including:
according to a molecular structure of a substance to be tested, determining substance features of the substance to be tested;
performing feature extraction on a diseased cell of a target category to obtain at least one cell feature of the diseased cell; and
according to the substance features and the at least one cell feature, predicting a response result of the substance to be tested against the diseased cell.

### Claim 13 (depends on 12)

The electronic device of claim 12, wherein the processor is further configured to:
according to the molecular structure of the substance to be tested, construct a structure feature map of the substance to be tested, wherein the structure feature map includes at least two nodes and lines between the nodes, each node represents an atom in the molecular structure, and each line represents an atomic bond in the molecular structure; and
according to the structure feature map, determine the substance features of the substance to be tested.

### Claim 14 (depends on 13)

The electronic device of claim 13, wherein the processor is further configured to:
according to the structure feature map, obtain a first adjacent matrix and a first feature matrix of the substance to be tested, wherein the first adjacent matrix represents neighbor relationships between atoms of the substance to be tested, and the first feature matrix represents attribute data of each atom of the substance to be tested; and
according to the first adjacent matrix and the first feature matrix, obtain the substance features of the substance to be tested.

### Claim 15 (depends on 14)

The electronic device of claim 14, wherein the processor is further configured to:
construct a complementary matrix of the first adjacent matrix according to a preset input dimension and a dimension of the first adjacent matrix, and construct a complementary matrix of the first feature matrix according to the preset input dimension and a dimension of the first feature matrix;
concatenate the first adjacent matrix and the complementary matrix of the first adjacent matrix to obtain a second adjacent matrix with the preset input dimension, and concatenate the first feature matrix and the complementary matrix of the first feature matrix to obtain a second feature matrix with the preset input dimension; and
perform graph convolution processing on the second adjacent matrix and the second feature matrix to obtain the substance features of the substance to be tested.

### Claim 16 (depends on 15)

The electronic device of claim 15, wherein, in the second adjacent matrix, the first adjacent matrix has no adjacent relationship with the complementary matrix of the first adjacent matrix.

### Claim 17 (depends on 15)

The electronic device of claim 15, wherein the processor is further configured to:
according to the first adjacent matrix and the complementary matrix of the first adjacent matrix, construct a first conjunction matrix, wherein elements in the first conjunction matrix are all preset values;
connect the first adjacent matrix and the complementary matrix of the first adjacent matrix through the first conjunction matrix to obtain the second adjacent matrix with the preset input dimension; and
connect the first feature matrix and the complementary matrix of the first feature matrix to obtain the second feature matrix with the preset input dimension.

### Claim 18 (depends on 12)

The electronic device of claim 12, wherein the processor is further configured to perform at least one of:
performing feature extraction on genomic mutation of the diseased cell to obtain a genome feature of the diseased cell;
performing feature extraction on gene expression of the diseased cell to obtain a transcriptome feature of the diseased cell; or
performing feature extraction on Deoxyribonucleic Acid (DNA) methylation data of the diseased cell to obtain an epigenome feature of the diseased cell.

### Claim 19 (depends on 12)

The electronic device of claim 12, wherein the processor is further configured to:
concatenate the substance features and the at least one cell feature to obtain a combined feature after concatenation; and
perform convolution processing on the combined feature to obtain a predicted response result of the substance to be tested against the diseased cell.

### Claim 20

A non-transitory computer-readable storage medium, having stored thereon computer program instructions that, when executed by a processor of an electronic device, cause the processor to perform a prediction method comprising:
according to a molecular structure of a substance to be tested, determining substance features of the substance to be tested;
performing feature extraction on a diseased cell of a target category to obtain at least one cell feature of the diseased cell; and
according to the substance features and the at least one cell feature, predicting a response result of the substance to be tested against the diseased cell.
