Washington University in St. Louis

Cortical surface registration network that learns which folds predict functional organization, using multi-modal supervision during training only.

Cell type annotation model mapping human single-cell and spatial transcriptomes onto one hierarchical typology of 381 types across 23 tissues.

Single-cell language model that prepends biomedical knowledge-graph tokens to cell sentences, grounding cell type annotation in pathway structure.

DNA language model whose tokenizer partitions the genome by evolutionary conservation, keeping transcription factor motifs intact as single tokens.

Protein conformational sampling framework that steers a retrained OpenFold with diverse secondary-structure predictions to recover alternative states.

De novo protein design model that co-generates sidechains, backbone, and sequence in one flow-matching process instead of backbone only.

Multimodal graph foundation model fusing single-cell expression, biomedical text, and signaling networks, pretrained on ~80M sc/snRNA-seq profiles.

Disordered protein ensemble prediction from sequence, generating hundreds of conformers in seconds via latent diffusion over distance maps.

Instance segmentation for nervous-system tissue, trained only on biophysical simulations and applied to real brain, spinal cord and nerve sections.

Intrinsic disorder prediction from protein sequence at proteome scale, distilling consensus disorder scores and AlphaFold2 pLDDT into one network.

EHR foundation model that writes patient histories as token sequences carrying explicit visit and day-interval tokens, invertible back to OMOP tables.

Per-residue AlphaFold2 pLDDT confidence regressed from sequence by a bidirectional LSTM, with no structure prediction and no database lookup.