GANs for latent space visualizations
The disclosure provides a method of analyzing a patient. The method includes receiving a plurality of latent space representations of patient data including genetic data associated with a plurality of patients, providing each of the plurality of latent space representations to a trained model, receiving a plurality of images from the trained model, each image included in the plurality of images being associated with a patient included in the plurality of patients, grouping at least a portion of the plurality of images into a plurality of groups, and displaying the plurality of groups to at least one user.
1. A method of analyzing a subject, the method comprising:
receiving first genetic data comprising at least one of RNA data or DNA data associated with a subject;
applying a machine learning model or a dimensionality reduction algorithm to the first genetic data to generate second genetic data having a lower dimensionality than the first genetic data;
providing the second genetic data to a trained model;
receiving a subject image generated by the trained model using the second genetic data;
identifying a pattern in the subject image that is present in at least one other image in a set of images generated by the trained model;
providing the subject image to an analysis engine;
receiving subject information from the analysis engine; and
causing the subject information to be output to at least one of a medical practitioner or a memory.
2. The method of claim 1 , wherein the first genetic data comprises at least one thousand RNA expression levels, and wherein the second genetic data comprises no more than one hundred dimensions of values.
3. The method of claim 1 , wherein applying a machine learning model or a dimensionality reduction algorithm to the first genetic data to generate second genetic data comprises:
providing the first genetic data to a trained autoencoder; and
receiving the second genetic data from the trained autoencoder.
4. The method of claim 1 , wherein the trained model comprises a neural network.
5. The method of claim 4 , wherein the neural network is a U-Net.
6. The method of claim 1 , wherein the trained model comprises a generator comprising a neural network, the generator being previously trained by training a training model comprising the generator and a discriminator based on a set of latent space representations and a set of images.
7. The method of claim 1 , wherein applying a machine learning model or a dimensionality reduction algorithm to the first genetic data to generate second genetic data comprises: applying one or more of principal component analysis (PCA) or t-distributed stochastic neighbor embedding to the first genetic data.
8. The method of claim 6 , wherein each latent space representation included in the set of latent space representations is generated by a trained autoencoder, and wherein the second genetic data comprises a latent space representation generated by the trained autoencoder.
9. The method of claim 6 , wherein the training the training model comprises:
providing a latent space representation included in the set of latent space representations to the generator;
receiving a generated image from the generator;
providing the generated image and an image included in the set of images to the discriminator; and
updating weights included in the generator and the discriminator based on the generated image and the image included in the set of images to the discriminator.
10. The method of claim 9 , wherein the set of images comprises object images.
11. The method of claim 10 , wherein the set of images comprises object images comprises images of at least one of an airplane, a bird, a car, a cat, a deer, a dog, a horse, a monkey, a ship, or a truck.
12. The method of claim 6 , wherein the set of images comprises cell images.
13. The method of claim 12 , wherein the cell images comprise slide images.
14. The method of claim 1 , wherein the subject information comprises information from a cohort of patients subjects.
15. The method of claim 1 , wherein the subject information comprises at least one potential treatment for the subject.
16. The method of claim 1 , wherein the identifying the pattern in the subject image comprises clustering the subject image with a subset of the set of images generated by the trained model.
17. The method of claim 16 , wherein the analysis engine is configured to identify at least one diagnostic metric between the subject image and a group of patients subjects associated with the subset of the set of images.
18. The method of claim 17 , wherein the at least one diagnostic metric comprises at least one of a gene variant present in the subject and at least a portion of the group of patients subjects, a disease present in the subject and at least a portion of the group of patients subjects, a preexisting condition present in the subject and at least a portion of the group of patients subjects, a treatment provided to at least a portion of the group of patients subjects, or a trial participated in by at least a portion of the group of patients subjects.
19. A system for analyzing a subject, the system comprising:
at least one memory; and
at least one processor coupled to the at least one memory, the system configured to cause the at least one processor to execute instructions stored in the at least one memory to:
receive first genetic data comprising at least one of RNA data or DNA data associated with a subject;
generate apply a machine learning model or a dimensionality reduction algorithm to the first genetic data to generate second genetic data having a lower dimensionality than the first genetic data;
provide the second genetic data to a trained model;
receive a subject image generated by the trained model using the second genetic data;
identify a pattern in the subject image that is present in at least one other image in a set of images generated by the trained model;
provide the subject image to an analysis engine;
receive subject information from the analysis engine; and
cause the subject information to be output to at least one of a medical practitioner or a memory.
20. A computer program product for analyzing a subject, the computer program product comprising instructions stored on a non-transitory computer readable medium to cause at least one processor to:
receive first genetic data comprising at least one of RNA data or DNA data associated with a subject;
apply a machine learning model or a dimensionality reduction algorithm to the first genetic data to generate second genetic data having a lower dimensionality than the first genetic data;
provide the second genetic data to a trained model;
receive a subject image generated by the trained model using the second genetic data;
identify a pattern in the subject image that is present in at least one other image in a set of images generated by the trained model;
provide the subject image to an analysis engine;
receive subject information from the analysis engine; and
cause the subject information to be output to at least one of a medical practitioner or a memory.