Johns Hopkins University

Reinforcement learning framework that fine-tunes the ProGen2-OAS antibody language model with GRPO to cut germline bias in generated sequences.

Protein-protein docking model adapting AlphaFold-Multimer with a docking module and flow-matching training to assemble subunits without MSAs.

Schrödinger-bridge diffusion model for virtual multiplex staining, translating routine H&E histology into multiplex immunohistochemistry images.

Yeast sequence-to-expression model pretrained as a masked DNA language model on 165 fungal genomes, predicting RNA-seq coverage and variant effects.

Gene Ontology prediction for malaria parasite proteins, trained on SAR supergroup structure embeddings with calibrated uncertainty and FDR control.

Modality-agnostic foundation model for human brain imaging that runs five core neuroimaging tasks across uncalibrated CT and MRI without retraining.

Single-cell type annotation pairing a CellMarker-derived knowledge graph with multi-agent LLM retrieval, generalizing across 11 tissue types.

Splice-site and variant-impact prediction from DNA sequence, with pretrained models for human, mouse, zebrafish, honey bee, and Arabidopsis.

Medical vision-language model trained with reinforcement learning for generalizable reasoning across eight imaging modalities and five question types.

Whole-body CT segmentation covering 235 fine-grained anatomies: 193 organs, 33 lymph node stations, and 9 lesion types from one unified network.

Whole-protein carbohydrate-binding classification from structure, for proteome-scale screening with graph convolutions over ESM2 embeddings.

Residue-level carbohydrate binding-site identification on protein structures, using equivariant graph convolutions over ESM2 per-residue embeddings.

4B-parameter generative genome foundation model trained on assembled environmental metagenomes for microbial representation and de novo DNA design.

Single-cell foundation model pretrained by federated learning, modeling expression as a cell-by-gene table rather than an ordered gene sentence.

Glucose forecasting from continuous glucose monitor streams over a two-hour horizon. Cuts one-hour rMSE 48.51% on OhioT1DM without training on it.

Diffusion model for protein-protein docking that unifies pose sampling and energy-based ranking, works without MSAs, and generalizes to new targets.

Mixture-of-Experts foundation model for medical image segmentation that generalizes across imaging modalities and clinical centers.

Organelle phenotyping model that scores how perturbations shift subcellular localization and morphology in confocal images of human neurons.

Abdominal CT segmentation model driven by CLIP text embeddings, covering 25 organs and 6 tumor types with zero-shot extension to new categories.