IP Library › Granted Patent US 11,574,703

Granted Patent B2

US 11,574,703 · App. 17/240,553 · Granted Feb 7, 2023

Method, apparatus, and computer-readable medium for efficiently optimizing a phenotype with a combination of a generative and a predictive model

Inventors: Eduardo Abeliuk (Oakland, CA); Andrés Igor Pérez Manríquez (Santiago, CL); Juan Andrés Ramírez Neilson (Santiago, CL); Diego Francisco Valenzuela Iturra (Santiago, CL)

Assignee: TESELAGEN BIOTECHNOLOGY INC.

G16B20/20 G06N3/0454 G06N3/08 G16B40/00

View Patent ↗

Loading inventors, assignments & file history…

Monitor This Case

Get email alerts when status or documents change.

Order Certified Copies

Most orders are placed with the USPTO same day — all within 24 business hours.

Order via The Patent Place →

Pre-filled with this patent's details

Quick Facts

Patent No.: US 11,574,703
App. No.: 17/240,553
Granted: Feb 7, 2023
Kind: B2

External Links

USPTO Patent Center↗ Full Patent PDF↗

Abstract

A method, apparatus, and computer-readable medium for efficiently optimizing a phenotype with a combination of a generative and a predictive model, training a phenotype prediction model based on experiential genotype vectors, training a genotype generation model based on sample genotype vectors, generating new genotype vectors, applying the phenotype prediction model to the new genotype vectors to generate scores, determining result genotypes based on a ranking of the available genotypes according to the scores, and generating a result based on the result genotypes, the result indicating one or more genetic constructs for testing.

Claims (134)

1. A method executed by one or more computing devices for efficiently optimizing a phenotype with a combination of a generative and a predictive model, the method comprising:

training, by at least one of the one or more computing devices, a phenotype prediction model based at least in part on the plurality of experiential genotype vectors, corresponding phenotype information, and one or more constraints, the phenotype prediction model comprising a surrogate model;

training, by at least one of the one or more computing devices, a genotype generation model based at least in part on a plurality of sample genotype vectors, the genotype generation model being configured to generate new genotype vectors;

generating, by at least one of the one or more computing devices, a plurality of new genotype vectors with the genotype generation model;

applying, by at least one of the one or more computing devices, the phenotype prediction model to the plurality of new genotype vectors to generate a plurality of scores, the phenotype prediction model being configured to predict one or more phenotypic attributes of the new genotype vectors;

determining, by at least one of the one or more computing devices, a plurality of result genotypes based at least in part on a ranking of the plurality of new genotype vectors according to the plurality of scores; and

generating, by at least one of the one or more computing devices, a result based at least in part on the plurality of result genotypes, the result indicating one or more genetic constructs for testing.

2. The method of claim 1 , further comprising:

receiving, by at least one of the one or more computing devices, one or more constraints, the one or more constraints comprising a plurality of desired phenotypic attributes; and

encoding, by at least one of the one or more computing devices, genotype information corresponding to the one or more constraints in a plurality of experimental data points as the plurality of experiential genotype vectors, the plurality of experimental data points comprising the genotype information and phenotype information corresponding to the genotype information.

3. The method of claim 2 , wherein encoding genotype information in a plurality of experimental data points corresponding to the one or more constraints as a plurality of experiential genotype vectors comprises:

identifying the plurality of experimental data points in a database of experimental data points based at least in part on at least one desired phenotypic attribute in the plurality of desired phenotypic attributes; and

encoding genotypes associated with the identified plurality of experimental data points as the plurality of experiential genotype vectors.

4. The method of claim 1 , further comprising:

encoding, by at least one of the one or more computing devices, genotype information in a plurality of sample genotypes of a sample database as the plurality of sample genotype vectors.

5. The method of claim 1 , wherein the phenotype prediction model is a surrogate model and wherein training a phenotype prediction model based at least in part on the plurality of experiential genotype vectors, the phenotype information, and the one or more constraints comprises:

determining, by at least one of the one or more computing devices, one or more meta-parameters for the phenotype prediction model, the one or more meta-parameters being configured to maximize accuracy of the phenotype prediction model;

determining an objective function based at least in part on the plurality of desired phenotypic attributes; and

iteratively adjusting the objective function by repeatedly selecting one or more experiential genotype vectors in the plurality of experiential genotype vectors that maximize an acquisition function of the phenotype prediction model and updating the objective function based at least in part on one or more experimentally-determined phenotypic attributes corresponding to the one or more experiential genotype vectors.

6. The method of claim 5 , wherein applying the phenotype prediction model to the plurality of new genotype vectors to generate a plurality of scores comprises:

applying the objective function to the plurality of new genotype vectors to generate a plurality of prediction scores corresponding to the plurality of new genotype vectors.

7. The method of claim 5 , wherein applying the phenotype prediction model to the plurality of new genotype vectors to generate a plurality of scores comprises:

applying an acquisition function of the phenotype prediction model to the plurality of new genotype vectors to generate a plurality of acquisition scores corresponding to the plurality of new genotype vectors.

8. The method of claim 7 , wherein generating a result based at least in part on the plurality of result genotypes comprises:

filtering the plurality of result genotype vectors to remove one or more first result genotype vectors corresponding to one or more categories of genotypes having genotype vectors with acquisition scores below acquisition scores of genotype vectors in other categories of genotypes;

selecting a plurality of filtered genotype vectors from the filtered plurality of result genotype vectors, the selected plurality of filtered genotype vectors corresponding to one or more additional categories of genotypes having genotype vectors with acquisition scores above acquisition scores of genotype vectors in other categories of genotypes;

determining a plurality of aggregate acquisition scores corresponding to a plurality of combinations of genotype vectors in the selected plurality of filtered genotype vectors;

ranking the plurality of combinations of genotype vectors according to the plurality of aggregate acquisition scores; and

selecting one or more top-ranked combinations of genotype vectors as the result, wherein each combination of genotype vectors corresponds to two or more genetic constructs for testing.

9. The method of claim 1 , wherein training a genotype generation model based at least in part on a plurality of sample genotype vectors, the genotype generation model being configured to generate new genotype vectors comprises:

storing a generator model function having a plurality of trainable generator parameters that is configured to mimic the distribution of the plurality of sample genotype vectors;

storing a discriminator model function having a plurality of trainable discriminator parameters that is configured to estimate a probability that a data sample comes from the plurality of sample genotype vectors instead of from the generator model function;

store a minimax objective function that is configured to be minimized by the generator model function and maximized by the discriminator model function; and

concurrently training both the generator model function and the discriminator model function with the plurality of sample genotype vectors until the minimax objective function converges to a saddle point.

10. The method claim 9 , wherein concurrently training both the generator model function and the discriminator model function with the plurality of sample genotype vectors until the minimax objective function converges to a saddle point comprises:

repeatedly sampling one or more sample genotype vectors from the plurality of sample genotype vectors;

repeatedly generating one or more generated genotype vectors with generator model function; and

iteratively applying the discriminator model function to the one or more sample genotype vectors and the one or more generated genotype vectors until the discriminator model function cannot distinguish between the one or more sample genotype vectors and the one or more generated genotype vectors, wherein application of the discriminator model function alternates between the one or more sample genotype vectors and the one or more generated genotype vectors.

11. The method of claim 9 , wherein generating a plurality of new genotype vectors with the genotype generation model comprises:

storing one or more parameters, the one or more parameters comprising a batch size and a selection rate; and

generating a set of new genotype vectors with the generator model function, the size of the set of new genotype vectors being determined based at least in part on the batch size and the selection rate.

12. The method of claim 1 , further comprising:

determining phenotype information corresponding to the one or more result genotype vectors in the plurality of result genotype vectors; and

re-training, by at least one of the one or more computing device, the phenotype prediction model based at least in part on the one or more result genotype vectors, the corresponding phenotype information, and the one or more constraints.

13. A apparatus executed by one or more computing devices for efficiently optimizing a phenotype with a combination of a generative and a predictive model, the apparatus comprising:

one or more processors; and

one or more memories operatively coupled to at least one of the one or more processors and having instructions stored thereon that, when executed by at least one of the one or more processors, cause at least one of the one or more processors to:

train a phenotype prediction model based at least in part on the plurality of experiential genotype vectors, corresponding phenotype information, and one or more constraints, the phenotype prediction model comprising a surrogate model;

train a genotype generation model based at least in part on a plurality of sample genotype vectors, the genotype generation model being configured to generate new genotype vectors;

generate a plurality of new genotype vectors with the genotype generation model;

apply the phenotype prediction model to the plurality of new genotype vectors to generate a plurality of scores, the phenotype prediction model being configured to predict one or more phenotypic attributes of the new genotype vectors;

determine a plurality of result genotypes based at least in part on a ranking of the plurality of new genotype vectors according to the plurality of scores; and

generate a result based at least in part on the plurality of result genotypes, the result indicating one or more genetic constructs for testing.

14. The apparatus of claim 13 , wherein at least one of the one or more memories has further instructions stored thereon that, when executed by at least one of the one or more processors, cause at least one of the one or more processors to:

receive one or more constraints, the one or more constraints comprising a plurality of desired phenotypic attributes; and

encode genotype information corresponding to the one or more constraints in a plurality of experimental data points as the plurality of experiential genotype vectors, the plurality of experimental data points comprising the genotype information and phenotype information corresponding to the genotype information.

15. The apparatus of claim 14 , wherein the instructions that, when executed by at least one of the one or more processors, cause at least one of the one or more processors to encode genotype information in a plurality of experimental data points corresponding to the one or more constraints as a plurality of experiential genotype vectors further cause at least one of the one or more processors to:

identify the plurality of experimental data points in a database of experimental data points based at least in part on at least one desired phenotypic attribute in the plurality of desired phenotypic attributes; and

encode genotypes associated with the identified plurality of experimental data points as the plurality of experiential genotype vectors.

16. The apparatus of claim 13 , wherein at least one of the one or more memories has further instructions stored thereon that, when executed by at least one of the one or more processors, cause at least one of the one or more processors to:

encode genotype information in a plurality of sample genotypes of a sample database as the plurality of sample genotype vectors.

17. The apparatus of claim 13 , wherein the phenotype prediction model is a surrogate model and wherein the instructions that, when executed by at least one of the one or more processors, cause at least one of the one or more processors to train a phenotype prediction model based at least in part on the plurality of experiential genotype vectors, the phenotype information, and the one or more constraints further cause at least one of the one or more processors to:

determine one or more meta-parameters for the phenotype prediction model, the one or more meta-parameters being configured to maximize accuracy of the phenotype prediction model;

determine an objective function based at least in part on the plurality of desired phenotypic attributes; and

iteratively adjust the objective function by repeatedly selecting one or more experiential genotype vectors in the plurality of experiential genotype vectors that maximize an acquisition function of the phenotype prediction model and updating the objective function based at least in part on one or more experimentally-determined phenotypic attributes corresponding to the one or more experiential genotype vectors.

18. The apparatus of claim 17 , wherein the instructions that, when executed by at least one of the one or more processors, cause at least one of the one or more processors to apply the phenotype prediction model to the plurality of new genotype vectors to generate a plurality of scores further cause at least one of the one or more processors to:

apply the objective function to the plurality of new genotype vectors to generate a plurality of prediction scores corresponding to the plurality of new genotype vectors.

19. The apparatus of claim 17 , wherein the instructions that, when executed by at least one of the one or more processors, cause at least one of the one or more processors to apply the phenotype prediction model to the plurality of new genotype vectors to generate a plurality of scores further cause at least one of the one or more processors to:

apply an acquisition function of the phenotype prediction model to the plurality of new genotype vectors to generate a plurality of acquisition scores corresponding to the plurality of new genotype vectors.

20. The apparatus of claim 19 , wherein the instructions that, when executed by at least one of the one or more processors, cause at least one of the one or more processors to generate a result based at least in part on the plurality of result genotypes further cause at least one of the one or more processors to:

filter the plurality of result genotype vectors to remove one or more first result genotype vectors corresponding to one or more categories of genotypes having genotype vectors with acquisition scores below acquisition scores of genotype vectors in other categories of genotypes;

select a plurality of filtered genotype vectors from the filtered plurality of result genotype vectors, the selected plurality of filtered genotype vectors corresponding to one or more additional categories of genotypes having genotype vectors with acquisition scores above acquisition scores of genotype vectors in other categories of genotypes;

determine a plurality of aggregate acquisition scores corresponding to a plurality of combinations of genotype vectors in the selected plurality of filtered genotype vectors;

rank the plurality of combinations of genotype vectors according to the plurality of aggregate acquisition scores; and

select one or more top-ranked combinations of genotype vectors as the result, wherein each combination of genotype vectors corresponds to two or more genetic constructs for testing.

21. The apparatus of claim 13 , wherein the instructions that, when executed by at least one of the one or more processors, cause at least one of the one or more processors to train a genotype generation model based at least in part on a plurality of sample genotype vectors, the genotype generation model being configured to generate new genotype vectors further cause at least one of the one or more processors to:

store a generator model function having a plurality of trainable generator parameters that is configured to mimic the distribution of the plurality of sample genotype vectors;

store a discriminator model function having a plurality of trainable discriminator parameters that is configured to estimate a probability that a data sample comes from the plurality of sample genotype vectors instead of from the generator model function;

store a minimax objective function that is configured to be minimized by the generator model function and maximized by the discriminator model function; and

concurrently train both the generator model function and the discriminator model function with the plurality of sample genotype vectors until the minimax objective function converges to a saddle point.

22. The apparatus claim 21 , wherein the instructions that, when executed by at least one of the one or more processors, cause at least one of the one or more processors to concurrently train both the generator model function and the discriminator model function with the plurality of sample genotype vectors until the minimax objective function converges to a saddle point further cause at least one of the one or more processors to:

repeatedly sample one or more sample genotype vectors from the plurality of sample genotype vectors;

repeatedly generate one or more generated genotype vectors with generator model function; and

iteratively apply the discriminator model function to the one or more sample genotype vectors and the one or more generated genotype vectors until the discriminator model function cannot distinguish between the one or more sample genotype vectors and the one or more generated genotype vectors, wherein application of the discriminator model function alternates between the one or more sample genotype vectors and the one or more generated genotype vectors.

23. The apparatus of claim 21 , wherein the instructions that, when executed by at least one of the one or more processors, cause at least one of the one or more processors to generate a plurality of new genotype vectors with the genotype generation model further cause at least one of the one or more processors to:

store one or more parameters, the one or more parameters comprising a batch size and a selection rate; and

generate a set of new genotype vectors with the generator model function, the size of the set of new genotype vectors being determined based at least in part on the batch size and the selection rate.

24. The apparatus of claim 13 , wherein at least one of the one or more memories has further instructions stored thereon that, when executed by at least one of the one or more processors, cause at least one of the one or more processors to:

determine phenotype information corresponding to the one or more result genotype vectors in the plurality of result genotype vectors; and

re-train the phenotype prediction model based at least in part on the one or more result genotype vectors, the corresponding phenotype information, and the one or more constraints.

25. At least one non-transitory computer-readable medium storing computer-readable instructions for efficiently optimizing a phenotype with a combination of a generative and a predictive model that, when executed by one or more computing devices, cause at least one of the one or more computing devices to:

train a genotype generation model based at least in part on a plurality of sample genotype vectors, the genotype generation model being configured to generate new genotype vectors;

generate a plurality of new genotype vectors with the genotype generation model;

determine a plurality of result genotypes based at least in part on a ranking of the plurality of new genotype vectors according to the plurality of scores; and

generate a result based at least in part on the plurality of result genotypes, the result indicating one or more genetic constructs for testing.

26. The at least one non-transitory computer-readable medium of claim 25 , further storing computer-readable instructions that, when executed by at least one of the one or more computing devices, cause at least one of the one or more computing devices to:

receive one or more constraints, the one or more constraints comprising a plurality of desired phenotypic attributes; and

27. The at least one non-transitory computer-readable medium of claim 26 , wherein the instructions that, when executed by at least one of the one or more computing devices, cause at least one of the one or more computing devices to encode genotype information in a plurality of experimental data points corresponding to the one or more constraints as a plurality of experiential genotype vectors further cause at least one of the one or more computing devices to:

encode genotypes associated with the identified plurality of experimental data points as the plurality of experiential genotype vectors.

28. The at least one non-transitory computer-readable medium of claim 25 , further storing computer-readable instructions that, when executed by at least one of the one or more computing devices, cause at least one of the one or more computing devices to:

encode genotype information in a plurality of sample genotypes of a sample database as the plurality of sample genotype vectors.

29. The at least one non-transitory computer-readable medium of claim 25 , wherein the phenotype prediction model is a surrogate model and wherein the instructions that, when executed by at least one of the one or more computing devices, cause at least one of the one or more computing devices to train a phenotype prediction model based at least in part on the plurality of experiential genotype vectors, the phenotype information, and the one or more constraints further cause at least one of the one or more computing devices to:

determine one or more meta-parameters for the phenotype prediction model, the one or more meta-parameters being configured to maximize accuracy of the phenotype prediction model;

determine an objective function based at least in part on the plurality of desired phenotypic attributes; and

30. The at least one non-transitory computer-readable medium of claim 29 , wherein the instructions that, when executed by at least one of the one or more computing devices, cause at least one of the one or more computing devices to apply the phenotype prediction model to the plurality of new genotype vectors to generate a plurality of scores further cause at least one of the one or more computing devices to:

apply the objective function to the plurality of new genotype vectors to generate a plurality of prediction scores corresponding to the plurality of new genotype vectors.

31. The at least one non-transitory computer-readable medium of claim 29 , wherein the instructions that, when executed by at least one of the one or more computing devices, cause at least one of the one or more computing devices to apply the phenotype prediction model to the plurality of new genotype vectors to generate a plurality of scores further cause at least one of the one or more computing devices to:

32. The at least one non-transitory computer-readable medium of claim 31 , wherein the instructions that, when executed by at least one of the one or more computing devices, cause at least one of the one or more computing devices to generate a result based at least in part on the plurality of result genotypes further cause at least one of the one or more computing devices to:

determine a plurality of aggregate acquisition scores corresponding to a plurality of combinations of genotype vectors in the selected plurality of filtered genotype vectors;

rank the plurality of combinations of genotype vectors according to the plurality of aggregate acquisition scores; and

select one or more top-ranked combinations of genotype vectors as the result, wherein each combination of genotype vectors corresponds to two or more genetic constructs for testing.

33. The at least one non-transitory computer-readable medium of claim 25 , wherein the instructions that, when executed by at least one of the one or more computing devices, cause at least one of the one or more computing devices to train a genotype generation model based at least in part on a plurality of sample genotype vectors, the genotype generation model being configured to generate new genotype vectors further cause at least one of the one or more computing devices to:

store a generator model function having a plurality of trainable generator parameters that is configured to mimic the distribution of the plurality of sample genotype vectors;

store a minimax objective function that is configured to be minimized by the generator model function and maximized by the discriminator model function; and

34. The at least one non-transitory computer-readable medium claim 33 , wherein the instructions that, when executed by at least one of the one or more computing devices, cause at least one of the one or more computing devices to concurrently train both the generator model function and the discriminator model function with the plurality of sample genotype vectors until the minimax objective function converges to a saddle point further cause at least one of the one or more computing devices to:

repeatedly sample one or more sample genotype vectors from the plurality of sample genotype vectors;

repeatedly generate one or more generated genotype vectors with generator model function; and

35. The at least one non-transitory computer-readable medium of claim 33 , wherein the instructions that, when executed by at least one of the one or more computing devices, cause at least one of the one or more computing devices to generate a plurality of new genotype vectors with the genotype generation model further cause at least one of the one or more computing devices to:

store one or more parameters, the one or more parameters comprising a batch size and a selection rate; and

generate a set of new genotype vectors with the generator model function, the size of the set of new genotype vectors being determined based at least in part on the batch size and the selection rate.

36. The at least one non-transitory computer-readable medium of claim 25 , further storing computer-readable instructions that, when executed by at least one of the one or more computing devices, cause at least one of the one or more computing devices to:

determine phenotype information corresponding to the one or more result genotype vectors in the plurality of result genotype vectors; and

re-train the phenotype prediction model based at least in part on the one or more result genotype vectors, the corresponding phenotype information, and the one or more constraints.

Assignments (1)

ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Dec 23, 2022

From: ABELIUK, EDUARDO; NEILSON, JUAN ANDRES RAMIREZ; MANRIQUEZ, ANDRES IGOR PEREZ; ITURRA, DIEGO FRANCISCO VALENZUELA

To: TESELAGEN BIOTECHNOLOGY INC.

Reel/Frame 062198/0872 →

Continuity (3)

Continuation In Part 16725642 · Dec 23, 2019

Provisional Application 63015140 · Apr 24, 2020

Related Publication 20210257049A1 · Aug 19, 2021

Cited By (4)

US 12,368,503 US 12,587,274 US 12,603,701 US 12,627,372