University of Pennsylvania

Cell-level pathology foundation model that types every nucleus on a routine H&E slide, supervised by paired Xenium spatial transcriptomics.

Residue-level prediction of ligand-binding and cryptic pockets in one model, reading frozen ESM-2 embeddings with no 3D structure at inference.

Masked discrete-diffusion model over millions of full-length mRNAs, steered by Monte Carlo tree search for joint codon optimization and UTR design.

Cell world model pretrained on a 2.4M-cell mouse embryonic atlas, predicting one-step transcriptional state transitions and perturbation response.

Self-supervised transformer for population genetics, pretrained on 1000 Genomes data, that detects positive selection via haplotype-wise attention.

Peptide developability predictor scoring solubility, permeability, toxicity, and binding from amino-acid sequences or chemically modified SMILES.

Kidney-specialized single-cell foundation model trained across four mammalian species for zero-shot cell-type annotation and batch integration.

De novo antibiotic design framework coupling a 6.4B-parameter protein language model with reinforcement learning to generate antimicrobial peptides.

Antimicrobial discovery model predicting compound potency against unseen bacterial strains and generating de novo antibiotics from pathogen genomes.

De novo design of heavy metal-binding peptides by classifier-guided diffusion over ESM-2 embeddings, with Cu and Zn binders validated in vitro.

MSA design model generating alignments from protein language model embeddings to improve folding accuracy on orphan and low-homology proteins.

Matryoshka sparse autoencoders scaled to ESM2-3B, exposing the features that drive protein structure prediction and steering ESMFold outputs.

Histopathology model predicting gene expression and DNA methylation from H&E slides across 23 cancer types, fusing FFPE and fresh-frozen predictors.

Multi-transformer model that predicts tissue-specific alternative splicing outcomes and generalizes zero-shot to unseen cellular conditions.

Antimicrobial peptide generator fine-tuned from ProGen2, trained against a frozen ESM-2 encoder's latent space as an approximate function checker.

Malate dehydrogenase sequence generator fine-tuned from ProGen2, with a latent-space distance term that lifted functional generations to 96.8%.

Drug-target affinity prediction pairing an SE(3)-equivariant GNN over 3D protein structure with a molecular GNN and residue-atom cross-attention.

Antimicrobial peptide optimization framework pairing a transformer VAE latent space with constrained Bayesian optimization against an MIC oracle.

Conditional denoising diffusion model that synthesizes chest radiographs from patient demographics and echocardiographic ventricular measurements.

Whole-genome somatic copy-number aberration prediction from bulk RNA-seq alone, with one pan-cancer model covering 33 tumor types.

Spatial transcriptomics prediction from H&E slides, inferring spot-level expression and tumor microenvironment composition in breast cancer.

RNA language model that predicts G-quadruplex formation and subtype from transcript sequence and scores how single-nucleotide variants alter folding.

Open-source, lightweight generalist vision-language foundation model for diverse biomedical imaging and text tasks.

Antimicrobial peptide generator running denoising diffusion in the continuous ESM-2 embedding space, validated in mouse infection models.

Spot detection for single-molecule RNA FISH and fluorescence microscopy, trained on a differentiable F1 approximation, needing no threshold tuning.