Utilizing masked autoencoder generative models to extract cellular response representation embeddings

A system comprising:

• at least one processor;
• and
• at least one non-transitory computer-readable storage medium storing instructions that, when executed by the at least one processor, cause the system to:
• generate a set of training cellular response representations from microscopy images of cells exposed to a set of perturbations;
• generate, utilizing a first machine learning model, a set of training cellular response representation embeddings from the set of training cellular response representations by embedding the set of training cellular response representations in a machine learning feature space as vector representations;
• generate perturbation significance values for the set of training cellular response representations relative to sampled subsets of the training cellular response representations by comparing, within the machine learning feature space, training cellular response representation embeddings from the set of training cellular response representation embeddings;
• filter the set of training cellular response representations utilizing the perturbation significance values corresponding to the set of training cellular response representation embeddings to identify a focused subset of the training cellular response representations for a staged training process;
• generate masked training cellular response representations from the focused subset of the training cellular response representations;
• and
• train a second machine learning model via the staged training process to generate cellular response representation embeddings for generating biological inferences in a drug discovery pipeline by:
• modifying parameters of the second machine learning model during a cellular response representation completion training stage utilizing the masked training cellular response representations corresponding to the focused subset of the training cellular response representations;
• identifying a first set of layers of the second machine learning model and a second set of layers of the second machine learning model;
• freezing the first set of layers of the second machine learning model during for a perturbation classification training stage to preserve learned parameters from the cellular response representation completion training stage;
• and
• selectively modifying the second set of layers of the second machine learning model during the perturbation classification training stage utilizing a perturbation classification task.

A non-transitory computer-readable medium storing instructions that, when executed by at least one processor, cause a computing device to:

• generate a set of training cellular response representations from transcriptomic profiles of cells exposed to a set of perturbations;
• generate, utilizing a first machine learning model, a set of training cellular response representation embeddings from the set of training cellular response representations by embedding the set of training cellular response representations in a machine learning feature space as vector representations;
• generate perturbation significance values for the set of training cellular response representations relative to sampled subsets of the training cellular response representations by comparing, within the machine learning feature space, training cellular response representation embeddings from the set of training cellular response representation embeddings;
• filter the set of training cellular response representations utilizing the perturbation significance values corresponding to the set of training cellular response representation embeddings to identify a focused subset of the training cellular response representations for a staged training process;
• generate masked training cellular response representations from the focused subset of the training cellular response representations;
• and
• train a second machine learning model via the staged training process to generate cellular response representation embeddings for generating biological inferences in a drug discovery pipeline by:
• modifying parameters of the second machine learning model during a cellular response representation completion training stage utilizing the masked training cellular response representations corresponding to the focused subset of the training cellular response representations;
• identifying a first set of layers of the second machine learning model and a second set of layers of the second machine learning model;
• freezing the first set of layers of the second machine learning model during a perturbation classification training stage to preserve learned parameters from the cellular response representation completion training stage;
• and
• selectively modifying the second set of layers of the second machine learning model during the perturbation classification training stage utilizing a perturbation classification task.