Granted Patent B2

US 12,282,851 · App. 17/842,247 · Granted Apr 22, 2025

Inventors: Aleksandr Aliper (Moscow, RU); Aleksandrs Zavoronkovs (Rockville, MD); Alexander Zhebrak (Moscow, RU); Artur Kadurin (Taipei, TW); Daniil Polykovskiy (Moscow, RU); Rim Shayakhmetov (Moscow, RU)

Assignee: INSILICO MEDICINE IP LIMITED

G06N3/08G06N3/088G06N7/01G10L19/24

Abstract

Claims (109)

1. A method, comprising:

providing a dataset having object data for an object and condition data for a condition;

providing a computing system having a deep neural network configured with a mutual information adversarial autoencoder;

processing the object data and condition data through the mutual information adversarial autoencoder of the deep neural network to obtain output data;

selecting a selected object based on the output data from the mutual information adversarial autoencoder of the deep neural network that satisfies a given condition;

comparing the output from the mutual information adversarial autoencoder with the object data and condition data;

calculating an object loss between generated object data of the output with input object data, for the selected object;

comparing the object loss to a threshold;

providing the selected object with the object loss less than the threshold;

generating the selected object with a decoder of the deep neural network to obtain a generated object that satisfies the given condition; and

providing the generated object that satisfies the given condition.

2. The method of claim 1 , further comprising:

comparing the output from the mutual information adversarial autoencoder with the object data and condition data;

calculating a condition loss between generated condition data of the output with input condition data;

comparing the condition loss to a threshold; and

selecting a generated condition that is less than threshold.

3. The method of claim 1 , further comprising:

comparing the output from the mutual information adversarial autoencoder with the object data and condition data;

selecting a generated object that is less than a threshold; and

selecting a generated condition that is less than a threshold.

4. The method of claim 1 , further comprising:

obtaining output from the mutual information adversarial autoencoder with the object data and condition data;

obtaining generated object-condition data from latent object data and latent object-condition data of the output;

comparing generated object-condition data with object-condition data; and

selecting generated object-condition data that is less than a threshold.

5. The method of claim 4 , further comprising:

obtaining output from the mutual information adversarial autoencoder with the object data and condition data;

obtaining generated condition-object data from latent condition data and latent condition-object data of the output;

comparing generated condition-object data with condition-object data; and

selecting generated condition-object data that is less than a threshold.

6. The method of claim 1 , further comprising:

processing the object data of the dataset with an object encoder;

processing the condition data of the dataset with a condition encoder;

processing latent object data from the object encoder to obtain generated object data with an object decoder;

processing the latent condition data from the condition encoder to obtain generated condition data with a condition decoder;

comparing the generated object data with the object data;

comparing the generated condition data with the condition data; and

determine a loss from generated data compared to input data.

7. The method of claim 6 , further comprising:

obtaining multidimensional vectors from generated data;

determining a generated data discriminator value;

sampling a reference distribution from a predefined multidimensional distribution;

determining a reference data discriminator value; and

determining a difference between the generated data discriminator value and the reference data discriminator value.

8. The method of claim 7 , further comprising performing an optimization protocol on a loss function based on the difference between the generated data discriminator value and the reference data discriminator value.

9. The method of claim 1 , further comprising:

processing the object data of the dataset with an object encoder;

processing the condition data of the dataset with a condition encoder;

processing latent object data from the object encoder to obtain generated object data with an object decoder;

processing the latent condition data from the condition encoder to obtain generated condition data with a condition decoder;

processing the latent object data and latent condition data to obtain a discriminator value;

selecting a selected object based on outcome in view of the discriminator value for a given condition; and

providing the selected object in a report with a recommendation for validation of a physical form of the object.

10. The method of claim 1 , further comprising obtaining a physical form of the selected object.

11. The method of claim 10 , further comprising validating the physical form of the object.

12. A computer program product comprising:

a non-transient, tangible memory device having computer-executable instructions that when executed by a processor, cause performance of a method comprising:

providing a dataset having object data for an object and condition data for a condition;

providing a computing system having a deep neural network configured with a mutual information adversarial autoencoder;

processing the object data and condition data through the mutual information adversarial autoencoder to obtain output data;

selecting a selected object based on the output data from the mutual information adversarial autoencoder of the deep neural network that satisfies a given condition;

comparing the output from the mutual information adversarial autoencoder with the object data and condition data;

calculating an object loss between generated object data of the output with input object data, for the selected object;

comparing the object loss to a threshold;

providing the selected object with the object loss less than the threshold;

generating the selected object with a decoder of the deep neural network; and

providing the generated object that satisfies the given condition.

13. The computer program product of claim 12 , the method further comprising:

comparing the output from the mutual information adversarial autoencoder with the object data and condition data;

calculating a condition loss between generated condition data of the output with input condition data;

comparing the condition loss to a threshold; and

selecting a generated condition that is less than threshold.

14. The computer program product of claim 12 , the method further comprising:

comparing the output from the mutual information adversarial autoencoder with the object data and condition data;

selecting a generated object that is less than a threshold; and

selecting a generated condition that is less than a threshold.

15. The computer program product of claim 12 , the method further comprising:

obtaining output from the mutual information adversarial autoencoder with the object data and condition data;

obtaining generated object-condition data from latent object data and latent object-condition data of the output;

comparing generated object-condition data with object-condition data; and

selecting generated object-condition data that is less than a threshold.

16. The computer program product of claim 14 , the method further comprising:

obtaining output from the mutual information adversarial autoencoder with the object data and condition data;

obtaining generated condition-object data from latent condition data and latent condition-object data of the output;

comparing generated condition-object data with condition-object data; and

selecting generated condition-object data that is less than a threshold.

17. The computer program product of claim 12 , the method further comprising:

processing the object data of the dataset with an object encoder;

processing the condition data of the dataset with a condition encoder;

processing latent object data from the object encoder to obtain generated object data with an object decoder;

processing the latent condition data from the condition encoder to obtain generated condition data with a condition decoder;

comparing the generated object data with the object data;

comparing the generated condition data with the condition data; and

determine a loss from generated data compared to input data.

18. The computer program product of claim 17 , the method further comprising:

obtaining multidimensional vectors from generated data;

determining a generated data discriminator value;

sampling a reference distribution from a predefined multidimensional distribution;

determining a reference data discriminator value; and

determining a difference between the generated data discriminator value and the reference data discriminator value.

19. The computer program product of claim 18 , the method further comprising performing an optimization protocol on a loss function based on the difference between the generated data discriminator value and the reference data discriminator value.

20. The computer program product of claim 12 , the method further comprising:

processing the object data of the dataset with an object encoder;

processing the condition data of the dataset with a condition encoder;

processing latent object data from the object encoder to obtain generated object data with an object decoder;

processing the latent condition data from the condition encoder to obtain generated condition data with a condition decoder;

processing the latent object data and latent condition data to obtain a discriminator value;

selecting a selected object based on outcome in view of the discriminator value for a given condition; and

providing the selected object in a report with a recommendation for validation of a physical form of the object.

Assignments (2)

Continuity (2)

References Cited (58)