University of Washington

Distilled whole-slide pathology foundation model pairing a 22M-parameter ViT-S tile encoder with a LongNet slide encoder for cohort-scale analysis.

Spatial proteomics prediction from routine H&E slides, generating 21-channel virtual multiplex immunofluorescence maps of the tumor microenvironment.

Multi-sequence spine MRI foundation model with DINOv3 encoders, supporting condition classification, pathology localization, and report generation.

Gene regulation model that conditions a pretrained DNA sequence embedding on CpG methylation to capture cell-type and allele-specific regulation.

Diffusion-based backbone generation and sequence design method for programmable asymmetric transmembrane beta-barrel nanopores.

Peptide ranking for targeted mass spectrometry, ordering a protein's precursors by expected DIA response to guide SRM and PRM assay design.

Discrete diffusion model for conditional antibody sequence design with germline-absorbing noising that focuses learning on somatic variation.

Liquid-biopsy deep learning model that infers transcriptome-wide tumor gene expression from standard-depth cell-free DNA whole-genome sequencing.

All-atom protein design diffusion model conditioned on ligands, nucleic acids, and other non-protein atoms, supporting enzyme and DNA binder design.

Atom-level diffusion model for de novo enzyme design that scaffolds arbitrary active-site geometries without specifying catalytic residue positions.

Gene expression prediction model combining DNA sequence with Hi-C contact maps to capture 3D chromatin looping behind cell-type-specific expression.

Multimodal foundation model that distills Evo 2 into a compact encoder guided by Hi-C data, predicting cell-type-specific 3D genome architecture.

Nucleic acid inverse-folding network that designs RNA sequences for a target 3D backbone and predicts protein-DNA binding specificity.

Histopathology model predicting homologous recombination deficiency from H&E slides in ovarian cancer, reaching 0.846 AUC and 0.938 specificity.

Multimodal diffusion transformer for de novo protein design, jointly generating sequence and structure conditioned on 465 Gene Ontology functions.

All-atom structure prediction for arbitrary biomolecular complexes of proteins, nucleic acids, and ligands, with code and weights under a BSD license.

Protein complex model quality assessment via DockQ-guided graph contrastive learning. CASP16 TMscore ranking loss of 0.123 versus 0.138 runner-up.

Hi-C resolution enhancement model combining a U²-Net with self-attention to recover TAD boundaries and chromatin loops from sparse contact maps.

General medical vision-language model trained with reinforcement learning to reason step by step over medical images for diagnosis and visual QA.

Protein sequence design model that represents small molecules, nucleotides, and metals at atomic resolution, enabling ligand-aware enzyme design.

Predicts haplotype-specific 3D genome organization and Hi-C contact maps from a single long-read Fiber-seq assay, using no DNA sequence as input.

Protein-ligand affinity foundation model that embeds pockets and ligands in one space, unifying virtual screening with hit-to-lead optimization.

Base-pair resolution sequence-to-activity CNN predicting ATAC-seq Tn5 insertion profiles and accessibility across 90 mouse immune cell types.

Liquid-liquid phase separation predictor that scores proteins and residues from sequence and designs phase-separating peptides by gradient descent.

Molecular docking framework that poses several ligands sharing one protein pocket at once, using their consistency to sharpen each prediction.

Cryo-EM reconstruction with neural radiance fields in Euclidean 3D space, separating conformational motion from compositional assembly states.

CRISPR-Cas PAM specificity prediction directly from Cas protein sequence, plus computational evolution of Cas9 variants toward a chosen PAM.

Structure-to-sequence protein design network inverting trRosetta, predicting coevolution features from a backbone to generate stable sequences.

Protein language model that generates paired heavy and light chain human antibodies from an antigen prompt, with binders validated in vitro.

Hi-C foundation model pretrained on 118 million contact submatrices, fine-tuned for loop detection, resolution enhancement and epigenomic prediction.

SMILES language model pretrained by editing: substructures are dropped and restored, giving fragment-level supervision for property prediction.

Target-conditioned peptide binder design model that samples hot-spot residues from an energy-based density, then extends fragments autoregressively.

Protein function prediction from 3D structure and sequence, assigning Gene Ontology terms with an ensemble built around rare long-tail terms.

Macrocyclic peptide binder design against protein targets, cyclizing a diffusion backbone generator's positional encoding so it closes rings.

Virtual multiplex immunofluorescence staining from H&E histopathology, imputing the expression and spatial localization of 50 protein biomarkers.

De novo enzyme design conditioned on the reaction to be catalysed: substrate and product SMILES in, catalytic pocket, enzyme, and docked complex out.

Single-cell model that ranks the genes driving a cell state transition, using a gene graph-enhanced manifold pretrained on 20 million cells.

Medical imaging embedding model spanning X-ray, CT, MRI, dermoscopy, OCT, fundus, ultrasound, histopathology and mammography in one encoder.

Enzyme catalytic pocket design conditioned on a reaction: substrate and product in, pocket backbone, sequence, and EC class out.

Structure-based mutational effect prediction from local atomic environments, scoring how substitutions change protein stability and binding affinity.

Bulk tumor transcriptome model ensembling hundreds of variational autoencoders into interpretable cancer-specific latent spaces for 18 cancers.

Deep network that predicts structures of full biological assemblies: proteins, nucleic acids, small molecules, metals, and covalent modifications.

De novo protein design diffusion model that generates backbone structures conditioned on binding targets, symmetry constraints, and functional motifs.

Protein structure prediction for monomers and complexes in one three-track network, scaling past 1000 residues without triangle attention.

AlphaFold fine-tuned on peptide-MHC and protein-peptide binding data for specificity prediction across MHC class I/II, PDZ, and SH3 domains.

Cyclic peptide structure prediction and de novo macrocycle design, by wrapping a frozen structure predictor's positional encoding into a ring.

Message passing neural network for fixed-backbone protein sequence design. Achieves 52.4% native sequence recovery, far surpassing Rosetta's 32.9%.

Protein-nucleic acid complex structure prediction from sequence, folding protein, DNA and RNA chains in one network with confidence estimates.

Structure accuracy estimation over atom graphs, scoring macrocyclic peptide and protein-DNA models that residue-level predictors cannot represent.

Full-atom protein model accuracy estimation, regressing per-atom lDDT with an SE(3)-transformer over a heavy-atom graph of the modeled structure.

Protein structure and complex prediction from sequence, in a three-track network that reasons over alignments, distances, and 3D coordinates at once.

Protein model accuracy estimation predicting per-residue lDDT plus signed residue-pair distance errors that become Rosetta refinement restraints.