SYSTEMS AND METHODS FOR MULTIMODAL GENERATIVE MACHINE LEARNING

• US 20190018933A1
• Filed: 01/13/2017
• Published: 01/17/2019
• Est. Priority Date: 01/15/2016
• Status: Active Application

• Current Assignee: Preferred Networks, Inc.
• Original Assignee: Preferred Networks, Inc.
• Inventors: Oono, Kenta, Clayton, Justin
• Application Number:
  US16/070,276
• Publication Number:
  US 20190018933A1
• Time in Patent Office: Days
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/047 Probabilistic or stochastic networks
• CPC Class Codes: G06N 7/08 using chaos models or non-linear system models
• CPC Class Codes: G16B 40/00 ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding
• CPC Class Codes: G16B 40/20 Supervised data analysis
• CPC Class Codes: G16C 20/70 Machine learning, data mining or chemometrics

Abstract

In various embodiments, the systems and methods described herein relate to multimodal generative models. The generative models may be trained using machine learning approaches, using training sets comprising chemical compounds and one or more of biological, chemical, genetic, visual, or clinical information of various data modalities that relate to the chemical compounds. Deep learning architectures may be used. In various embodiments, the generative models are used to generate chemical compounds that satisfy multiple desired characteristics of different categories.

Claims

1. A computer system comprising a multimodal generative model, the multimodal generative model comprising:

(a) a first level comprising n network modules, each having a plurality of layers of units; and

(b) a second level comprising m layers of units;

wherein the generative model is trained by inputting it training data comprising at least l different data modalities and wherein at least one data modality comprises chemical compound fingerprints.

2. The computer system of claim 1, wherein at least one of the n network modules comprises an undirected graph.

3. The computer system of claim 2, wherein the undirected graph comprises a restricted Boltzmann machine (RBM) or deep Boltzmann machine (DBM).

4. The computer system of claim 1, wherein at least one data modality comprises genetic information.

5. The computer system of claim 1, wherein at least one data modality comprises test results or image.

6. The computer system of claim 1, wherein a first layer of the second level is configured to receive input from a first inter-level layer of each of the n network modules.

7. The computer system of claim 6, wherein a second inter-level layer of each of the n network modules is configured to receive input from a second layer of the second level.

8. The computer system of claim 7, wherein the first layer of the second level and the second layer of the second level are the same.

9. The computer system of claim 7, wherein the first inter-level layer of a network module and the second inter-level layer of a network module are the same.

10. The computer system of claim 1, wherein n is at least 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30, 35, 40, 45, 50, 60, 70, 80, 90, or 100.

11. The computer system of claim 1, wherein m is at least 1, 2, 3, 4, or 5.

12. The computer system of claim 1, wherein l is at least 2, 3, 4, 5, 6, 7, 8, 9, or 10.

13. The computer system of claim 1, wherein the training data comprises a data type selected from the group consisting of genetic information, whole genome sequence, partial genome sequence, biomarker map, single nucleotide polymorphism (SNP), methylation pattern, structural information, translocation, deletion, substitution, inversion, insertion, viral sequence insertion, point mutation, single nucleotide insertion, single nucleotide deletion, single nucleotide substitution, microRNA sequence, microRNA mutation, microRNA expression level, chemical compound representation, fingerprint, bioassay result, gene expression level, mRNA expression level, protein expression level, small molecule production level, glycosylation, cell surface protein expression, cell surface peptide expression, change in genetic information, X-ray image, MR image, ultrasound image, CT image, photograph, micrograph, patient health history, patient demographic, patient self-report questionnaire, clinical notes, toxicity, cross-reactivity, pharmacokinetics, pharmacodynamics, bioavailability, and solubility.

14. The computer system of claim 1, wherein the generative model is configured to generate values for a chemical compound fingerprint upon input of genetic information and test results.

15. The computer system of claim 1, wherein the generative model is configured to generate values for genetic information upon input of chemical compound fingerprint and test result.

16. The computer system of claim 1, wherein the generative model is configured to generate values for test results upon input of chemical compound fingerprint and genetic information.

17. A method for training a generative model, comprising

(a) inputting it training data comprising at least l different data modalities, at least one data modality comprising chemical compound fingerprints;

wherein the generative model comprises

(i) a first level comprising n network modules, each having a plurality of layers of units; and

(ii) a second level comprising m layers of units.

18. A method of generating personalized drug prescription predictions, the method comprising:

(a) inputting to a generative model a value for genetic information and a fingerprint value for a chemical compound; and

(b) generating a value for test results;

wherein the generative model comprises

(i) a first level comprising n network modules, each having a plurality of layers of units; and

(ii) a second level comprising m layers of units;

wherein the generative model is trained by inputting it training data comprising at least l different data modalities, at least one data modality comprising chemical compound fingerprints, at least one data modality comprising test results, and at least one data modality comprising genetic information; and

wherein the likelihood of a patient having genetic information of the input value to have the generated test results upon administration of the chemical compound is greater than or equal to a threshold likelihood.

19. The method of claim 18, further comprising producing for the patient a prescription comprising the chemical compound.

20. The method of claim 18, wherein the threshold likelihood is at least 99%, 98%, 97%, 96%, 95%, 90%, 80%, 70%, 60%, 50%, 45%, 40%, 35%, 30%, 25%, 20%, 15%, 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, 1%, 0.5%, or 0.1%.

21. A method of personalized drug discovery, the method comprising:

(a) inputting to a generative model a test result value and a value for genetic information; and

(b) generating a fingerprint value for a chemical compound;

wherein the generative model comprises

(i) a first level comprising n network modules, each having a plurality of layers of units; and

(ii) a second level comprising m layers of units;

wherein the generative model is trained by inputting it training data comprising at least l different data modalities, at least one data modality comprising chemical compound fingerprints, at least one data modality comprising test results, and at least one data modality comprising genetic information; and

wherein the likelihood of a patient having genetic information of the input value to have the test results upon administration of the chemical compound is greater than or equal to a threshold likelihood.

22. The method of claim 21, wherein the threshold likelihood is at least 99%, 98%, 97%, 96%, 95%, 90%, 80%, 70%, 60%, 50%, 45%, 40%, 35%, 30%, 25%, 20%, 15%, 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, 1%, 0.5%, or 0.1%.

23. A method of identifying patient populations for a drug, the method comprising:

(a) inputting to a generative model a test result value and a fingerprint value for a chemical compound; and

(b) generating a value for genetic information;

wherein the generative model comprises

(i) a first level comprising n network modules, each having a plurality of layers of units; and

(ii) a second level comprising m layers of units;

wherein the generative model is trained by inputting it training data comprising at least l different data modalities, at least one data modality comprising chemical compound fingerprints, at least one data modality comprising test results, and at least one data modality comprising genetic information; and

wherein the likelihood of a patient having genetic information of the generated value to have the input test results upon administration of the chemical compound is greater than or equal to a threshold likelihood.

24. The method of claim 23, wherein the threshold likelihood is at least 99%, 98%, 97%, 96%, 95%, 90%, 80%, 70%, 60%, 50%, 45%, 40%, 35%, 30%, 25%, 20%, 15%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, 1%, 0.5%, or 0.1%.

25. The method of claim 23, further comprising:

conducting a clinical trial comprising a plurality of human subjects, wherein an administrator of the clinical trial has genetic information satisfying the generated value for genetic information for at least a threshold fraction of the plurality of human subjects.

26. The method of claim 25, wherein the threshold fraction is at least at least 99%, 98%, 97%, 96%, 95%, 90%, 80%, 70%, 60%, 50%, 45%, 40%, 35%, 30%, 25%, 20%, 15%, 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, 1%, 0.5%, or 0.1%.

27. A method of conducting a clinical trial for a chemical compound, the method comprising:

(a) administering to a plurality of human subjects the chemical compound,

wherein an administrator of the clinical trial has genetic information satisfying a generated value for genetic information for at least a threshold fraction of the plurality of human subjects and wherein the generated value for genetic information is generated according to the method of claim 23.

28. The method of claim 27, wherein the threshold fraction is at least at least 99%, 98%, 97%, 96%, 95%, 90%, 80%, 70%, 60%, 50%, 45%, 40%, 35%, 30%, 25%, 20%, 15%, 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, 1%, 0.5%, or 0.1%.