Tsinghua University

Volumetric cell segmentation for 3D fluorescence microscopy, pretrained on 5.1 TB of unlabeled volumes and generalizing across species and imaging.

Genomic foundation model that pairs 650 kb of gene-centered DNA with transcription factor activity to predict expression in unseen cell types.

Histopathology foundation model for uterine malignancies that orders whole-slide morphology into continuous, progression-associated tumor states.

Machine learning force field for all-atom protein dynamics, trained on 40 million DFT dipeptide conformations covering backbone and side-chain space.

De novo siRNA design conditioned on target mRNA context, generating 21-nt duplexes by masked discrete diffusion under efficacy guidance.

EEG foundation model pretrained to predict structured latent states rather than masked waveforms, reaching 52.94% frozen macro balanced accuracy.

Genomic foundation model on a Mamba, attention, and mixture-of-experts backbone, with 1M-token context for variant scoring and regulatory DNA design.

Calcium imaging foundation model for neural population forecasting and behavior decoding, with frozen-backbone transfer across three species.

Vision-language model for neuroblastoma pathology that reads H&E slides with their reports to grade tumors, infer biomarkers and stratify risk.

Protein-text foundation model placing amino acid sequences and natural language in one token space for protein understanding and de novo design.

Tri-modal foundation model unifying histology images, spatial transcriptomics, and language for zero-shot pathology and spatial biology reasoning.

Physics-informed generative foundation model for quantitative diffusion MRI that maps brain microstructure and adapts zero-shot to each participant.

Medical agent system for continuous care, combining a reasoning model with clinical tools for patient memory, evidence retrieval, and medical imaging.

Protein structure prediction and binder design in a single generative step, replacing AlphaFold3's iterative diffusion sampling with one forward pass.

Co-generative protein language model decoding sequence and structure tokens together from GO functional annotations for de novo protein design.

All-atom generative foundation model that designs small molecules, peptides, and nanobodies against a target binding site from a single checkpoint.

Molecular reasoning model built on DeepSeek-7B, using chain-of-thought and reinforcement learning for property prediction, generation, and reactions.

Hierarchical language model for atlas-level cell-type annotation of scATAC-seq data that annotates new query datasets without retraining.

Renal pathology foundation model whose learning unit is a whole detected glomerulus, pretrained on over a million of them across four biopsy stains.

Renal pathology foundation model self-supervised on a million kidney biopsy tiles spanning glomeruli, interstitium, and surrounding structures.

Protein language model that encodes sequences as discrete words from a learned vocabulary for zero-shot function inference and protein design.

Latent diffusion model that designs D-peptide binders against native L-protein targets, generalizing across chirality via axial vector features.

De novo peptide sequencing transformer that reads modified and unmodified peptides directly from tandem mass spectra without a reference database.

Equivariant diffusion model that converts peptide binders into drug-like small molecules, generating peptidomimetics inside the target protein pocket.

Multimodal LLM that tokenizes single cells into discrete VQ-VAE codebook tokens, letting one model reason jointly over transcriptomes and text.

Retrieval-augmented latent diffusion model for protein binder design, retrieving interfaces in a shared latent space across peptides and antibodies.

Hierarchical transformer for virus discovery in metagenomes, classifying viral genomes across taxonomic ranks and flagging candidate novel lineages.

Antibiotic resistance gene detection in metagenomes, pairing frozen ESM-1v embeddings with light classifier heads for drug class and mechanism.

Simplex diffusion model for discrete sequence generation, with released checkpoints for DNA enhancer design and de novo protein sequence design.

Protein-ligand binding model that embeds ligands, pockets, and sequences in hyperbolic space, unifying virtual screening and affinity ranking.

Drug perturbation model predicting post-treatment gene expression from a cell's baseline profile and compound structure, zero-shot on unseen drugs.

Spatial transcriptomics foundation model pairing gene-scale cell embeddings with an SE(2) Transformer over cell coordinates, pretrained on 88M cells.

Protein foundation model built on Bayesian Flow Networks, prompted with MSA profiles for structure- and function-preserving sequence design.

Self-supervised medical imaging foundation model pretrained on 3.3 million CT, X-ray, ultrasound, pathology, OCT, fundus, and dermoscopy images.

Spatial transcriptomics resolution enhancement from expression alone, using a tri-oriented Mamba encoder to predict expression between capture spots.

Structure-based virtual screening model that scores ligands against apo and predicted pockets, lifting blind-apo EF1% on DUD-E from 11.75 to 37.19.

mRNA optimization model that raises codon adaptation, tRNA adaptation, and folding stability at once while preserving the encoded protein sequence.

Cyclic peptide design conditioned on target protein structure, generating all four cyclization types via all-atom, all-bond harmonic SDE modeling.

Molecular dynamics emulator generating time-coarsened trajectories for small molecules, peptides, and proteins from one shared atomic representation.

Brain foundation model unifying EEG and MEG in a single encoder via a shared discrete tokenizer that transfers across sensor layouts and montages.

Structure-based drug design that schedules noise separately for 3D coordinates and 2D topology, reaching a 95.9% PoseBusters valid rate on CrossDock.

Spatial transcriptomics foundation model pretrained to generate a cell's expression profile from its neighbors, yielding zero-shot niche embeddings.

Hypergraph foundation model for brain disease diagnosis from resting-state fMRI, self-supervised on high-order connectivity among brain regions.

Structure-based drug design model pairing an autoregressive transformer for ligand graphs with a diffusion head for 3D binding-pose coordinates.

De novo binder design across small molecules, peptides, and antibodies from one geometric latent diffusion model over graphs of molecular blocks.

Protein inverse folding ensemble that fuses five pretrained sequence designers through a self-attention encoder, reaching 63.1% recovery on CATH4.2.

All-atom generative foundation model for biomolecular structure, unifying protein-ligand docking, structure-based drug design, and peptide design.

Generative foundation model that co-generates sequence and 3D coordinates for proteins, small molecules, and crystals under functional objectives.

Full-atom flow matching model that generates a ligand and the induced-fit holo pocket together, starting from an apo binding site.

PROTAC degradation prediction from molecular graphs of the target, linker, and E3 ligase, combining cross-attention with contrastive learning.

Histopathology foundation model pretrained on 300K whole-slide images across 20 tissue types and validated on 112 clinical-grade downstream tasks.

Prophage island detection in bacterial genomes and metagenome-assembled genomes, pairing a fine-tuned ESM-2 gene classifier with density clustering.

Latent diffusion model for single-cell multi-omics generation and modality translation, with gradient-based inference of gene regulatory networks.

Structure-based drug discovery transformer that handles protein-ligand docking and pocket-aware 3D molecule design in one pretrained model.

Multimodal ECG language model pairing a specialized signal encoder with a biomedical LLM for cardiovascular disease detection and question answering.

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

Protein function annotation model that parses sequences into residue clusters via community detection on ESM-2 attention, then maps them to GO terms.

Peptide binder design by mimicking the binding interface of a known receptor or antibody, generating all-atom peptides through latent diffusion.

Antibody CDR sequence-structure co-design by flow matching, starting from an informative structural prior rather than from Gaussian noise.

Multimodal vision-text foundation model for brain CT and MRI, pretrained on roughly 10 million image-report pairs to act as a clinical copilot.

Single-cell epigenomic foundation model that reads scATAC-seq as cell sentences of accessible cCREs, pretrained on about 5 million human cells.

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

Structure-based molecule optimization that steers a Bayesian flow network with property gradients over 3D coordinates and atom types at once.

Pocket-conditioned 3D ligand generator trained on its own predicted conformations, closing the train-inference gap that degrades diffusion sampling.

Protein language model that explains single-site mutation effects in natural language and proposes new mutants from free-text instructions.

Reprograms a frozen single-target diffusion model for dual-target drug design by composing SE(3)-equivariant messages across two aligned pockets.

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

Molecular docking and design foundation model that unifies structure-based drug design and peptide design at the atom level in one checkpoint.

Joint embedding space for common SNPs and free-text clinical concepts, aligned by contrastive learning over GWAS, biobank and knowledge-graph pairs.

Diffusion model for de novo protein backbone design that learns a 32-dimensional latent code for global fold geometry and generates conditioned on it.

Cis-regulatory element classifier that reads DNA sequence plus chromatin accessibility and loop tracks to label enhancers, silencers and insulators.

End-to-end framework predicting protein structure and mutational fitness from a single sequence, with five-fold faster inference than ESMFold.

Cell-free RNA language model for multi-cancer detection, classifying plasma samples straight from raw sequencing reads without gene annotation.

Scaling-law study of protein language models identifying compute-optimal training for causal and masked objectives on 939 million protein sequences.

Full-atom peptide binder design against a target pocket, generating backbone frames, side-chain torsions and residue types in one joint flow.

Structure-based drug design that generates 3D ligands for a protein pocket entirely in the continuous parameter space of a Bayesian flow network.

RNA language model that builds base-pairing constraints into self-attention, pretrained on 20.4 million sequences for structure and function tasks.

3D molecule generation that models atom coordinates and element types as distribution parameters updated by Bayesian inference, not by denoising.

All-atom 3D molecular foundation model pretrained across small molecules, proteins, and complexes with an E(3)-equivariant denoising objective.

Generative language model for single-cell transcriptomics with 368M parameters, unifying cell type annotation, batch integration, and cell generation.

Structure-based drug design diffusion model that re-extracts the essential binding subcomplex from a pocket at every step of 3D ligand generation.

Unified 100-billion-parameter protein language model combining autoencoding and autoregressive objectives for protein understanding and generation.

Vision-language model for annotation-free pathology localization, marking the finding a text prompt names in X-ray, histology and fundus images.

Diffusion model for synthesizing single-cell RNA-seq data, with guided generation of specific cell types, rare cells, and developmental trajectories.

Transformer model predicting context-specific epigenomic signals across cell types using DNA sequence and transcription factor activity profiles.

Structure-based drug design by SE(3)-equivariant diffusion over 3D atom coordinates and types, with the same frozen network scoring binding affinity.

Molecular graph transformer pretrained on 11 million unlabelled compounds, used as a frozen fingerprint source or fine-tuned for property prediction.