Learning useful biological representations from omics and imaging data with limited labels.

Connecting omics, images, and phenotypes while preserving shared and distinct biological information.

Modeling biological variation across continuous spatial coordinates and time to reconstruct tissue organization and dynamic processes.

Reconstructing disease phenotypes across tissue, cell, and molecular scales from noisy or incomplete measurements.

Linking pathology, spatial omics, and clinical phenotypes to identify molecular targets.

Learning from genetic, perturbation, and imaging data to predict compound function and support drug screening and repurposing.