Hugging Face

Meta AI's family of protein language models scaled to 15B parameters, paired with ESMFold for fast, alignment-free atomic-level structure prediction.

Sparse attention transformer that extends BERT to 8x longer sequences via random, local, and global attention, with genomic sequence applications.

Variational autoencoder pretrained on 74 million human single-cell transcriptomes from the CELLxGENE Census for batch correction and cell typing.

Biomedical vision-language assistant for question answering on radiology and pathology images, adapted from LLaVA on PubMed Central captions.

Computational pathology foundation model (ViT-L/16, DINOv2) pretrained on over 100 million H&E tiles from more than 100,000 whole-slide images.

Generative transformer pretrained on PubMed abstracts for biomedical text generation and mining, including relation extraction and question answering.

Promptable foundation model for universal medical image segmentation, fine-tuned from SAM on 1.57M image-mask pairs across 10 imaging modalities.

Suite of six protein language models, including ProtBERT and ProtT5, trained on up to 393 billion amino acids without multiple sequence alignments.

Single-cell foundation model pretrained on about 30 million human transcriptomes, using rank-value encoding for context-aware gene network inference.

Scientific large language model trained on 48 million papers, textbooks, and reference works to store, combine, and reason about scientific knowledge.

Histopathology vision-language foundation model pretrained on 1.17 million image-caption pairs with contrastive and captioning objectives.

Protein language model pretrained on UniRef90 with masked language modeling and Gene Ontology annotation prediction, at 16 million parameters.

Whole-slide histopathology foundation model pretrained on 1.3 billion image tiles from 171,189 clinical slides spanning 31 tissue types.

Self-supervised foundation model for retinal imaging, pretrained on 1.6 million unlabelled fundus and OCT scans to detect ocular and systemic disease.

Autoregressive protein language model based on GPT-2 that generates de novo protein sequences sampling unexplored regions of protein space.

Vision-language foundation model for pathology, fine-tuned from CLIP on 208,414 image-text pairs for zero-shot classification and image retrieval.

Biomedical vision-language model trained contrastively on 15M PubMed Central figure-caption pairs for zero-shot classification, retrieval, and VQA.

Multimodal medical vision-language model for few-shot visual question answering, learning new imaging tasks from in-context examples at inference.

Large-scale chemical language model trained on 1.1 billion SMILES strings using linear attention transformers for molecular property prediction.

Single-cell transcriptomics foundation model with 100 million parameters, pretrained on over 50 million human scRNA-seq profiles for cell embeddings.

Genomic foundation model built on the Hyena operator, processing DNA at single-nucleotide resolution with context windows up to 1 million tokens.

Multi-species genomic foundation model swapping k-mer tokenization for byte pair encoding, matching Nucleotide Transformer with 21x fewer parameters.

Open-source structure prediction model for proteins, nucleic acids, and small molecules, trained on public data to AlphaFold3-level accuracy.

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

Biomolecular structure prediction foundation model covering proteins, small molecules, DNA, RNA, and glycans in a single diffusion framework.

Medically tuned SigLIP encoder from Google that maps medical images and text into one embedding space for zero-shot classification and retrieval.

Open medical multimodal models from Google, built on Gemma 3 with a medically tuned SigLIP vision encoder for clinical text and image understanding.

Generative medical visual question answering model that pairs a vision encoder with a language model, trained on the 227k-pair PMC-VQA dataset.

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

Structure-aware protein language model pairing amino acid tokens with Foldseek 3Di structural states, outperforming ESM-2 across 10 downstream tasks.

Generative model that emulates protein equilibrium ensembles, sampling cryptic pockets and unfolded states far faster than molecular dynamics.

Multimodal generative protein language model reasoning jointly over protein sequence, structure, and function, trained at 98B parameters.

Medical-domain CLIP fine-tuned on radiology image-caption pairs from ROCO, serving as a drop-in visual encoder for medical visual question answering.

Genomic foundation model trained on 9.3 trillion DNA base pairs across all domains of life, with 40B parameters and a 1-million-token context.

Vision-language foundation model for precision oncology, pretrained on 50M pathology images and 1B text tokens via unified masked modeling.

Radiology foundation model that reads interleaved 2D and 3D scans with text for diagnosis, visual question answering, and report generation.

Protein conformational ensemble generator that fine-tunes AlphaFold 2 with flow matching, sampling protein dynamics beyond a single static structure.

SAM2-based foundation model that segments 2D and 3D medical images by treating volumes and image sets as video object tracking.

Regulatory genomics model predicting cell-type-specific RNA-seq coverage from DNA sequence, unifying transcription, splicing, and polyadenylation.

RNA foundation model pretrained on 23.7 million non-coding RNA sequences, producing embeddings for structure prediction, annotation, and RNA design.

Genomic foundation model with 7B parameters that models prokaryotic DNA, RNA, and protein at single-nucleotide resolution over a 131k-token context.

Histopathology foundation models: self-supervised vision transformers pretrained on millions of whole-slide images for tile-level feature extraction.

DNA foundation models from 500M to 2.5B parameters, trained on 3,200+ human genomes and 850 species for variant effect prediction.

Bidirectional, reverse-complement equivariant DNA language models built on Mamba state space models for long-range variant effect prediction.

End-to-end RNA 3D structure prediction from sequence alone, coupling the RNA-FM language model with an Invariant Point Attention structure module.

Open medical multimodal LLMs (7B and 34B) for visual question answering over radiology, pathology, and endoscopy images, trained on PubMedVision.

Generalist medical multimodal LLM for image understanding, visual question answering, and report generation across twelve-plus imaging modalities.

Generalist cell segmentation model pairing SAM's ViT-L encoder with Cellpose flow fields, outperforming average human annotators on its benchmark.

2B-parameter medical vision-language model that uses reinforcement learning to show interpretable reasoning for radiology visual question answering.

EEG foundation model for brain-computer interface decoding, factorizing self-attention into parallel spatial and temporal branches.

Fully 3D promptable segmentation foundation model for volumetric CT and MR, encoding whole volumes so anatomy can be segmented from one prompt point.

Multimodal large language model for 3D medical imaging that handles report generation, visual question answering, and segmentation on CT volumes.

Biomedical imaging foundation model that segments, detects, and recognizes structures across nine modalities from natural language prompts.

Multi-modal protein language model trained on sequences paired with biomedical text, enabling zero-shot function prediction and text-based retrieval.

DNA foundation model that predicts thousands of functional genomic tracks, from expression and splicing to chromatin, at single base-pair resolution.

Microsoft Research multimodal LLM for grounded chest X-ray report generation, localizing each described finding with bounding boxes on the image.

Transformer foundation model pretrained on 110M single-cell and spatial transcriptomics profiles, transferring spatial context to dissociated cells.

Protein language model that fuses Gene Ontology knowledge graphs with masked language modeling, improving protein function and interaction prediction.

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

3D vision-language foundation model for abdominal CT, pretrained on scans, radiology reports, and EHR codes for zero-shot interpretation.

RNA language model that learns base-level embeddings capturing sequence context and secondary structure, enabling fast structural alignment.

fMRI foundation model pretrained with masked autoencoding on roughly 6,700 hours of recordings for clinical prediction and network discovery.

Single-cell foundation model trained by metric learning to embed scRNA-seq profiles for cell type annotation and similarity search in cell atlases.

Promptable 3D foundation model for volumetric CT segmentation, covering over 200 anatomical categories through point, box, and free-text prompts.

Transformer language model for 5' UTR sequences that predicts mRNA translation efficiency, ribosome loading, and protein expression levels.

Virtual cell transformer that predicts how cells respond to genetic, chemical, or signaling perturbations, generalizing to unseen cellular contexts.

Medical vision-language model that unifies image comprehension and generation in one autoregressive transformer via heterogeneous LoRA adapters.

RNA language model trained on multiple sequence alignments of Rfam families, predicting secondary structure and solvent accessibility from homology.

Multimodal pathology assistant that answers questions about histology and cytology images, pairing the PathCLIP vision encoder with a Vicuna-13B LLM.

Text-guided protein design framework aligning language with sequences for text-conditioned generation, zero-shot editing, and property prediction.

Multi-task EEG foundation model that treats brain signals as a foreign language, pairing a text-aligned neural tokenizer with a GPT-2 backbone.

EEG foundation model that pairs a convolutional encoder with a GPT backbone, pretrained by masked-segment reconstruction for low-data BCI decoding.

Open transformer foundation model for 12-lead electrocardiograms, pretrained on 1.5 million unlabeled ECGs with a wav2vec 2.0 self-supervised recipe.

Medical vision-language model trained on the MedTrinity-25M dataset, answering questions and generating text about radiology and histology images.

Protein large language model adapted from LLaMA-2 that unifies sequence generation and superfamily classification in one 7B-parameter framework.

500M-parameter transformer model pretrained on intracranial SEEG recordings for neural signal forecasting, imputation, and seizure detection.

Histopathology foundation models pretrained with DINOv2 on over 1 million whole-slide images, released as Hibou-B and Hibou-L under Apache 2.0.

DNA language model for variant effect prediction across coding and non-coding regions, using whole-genome alignments of 100 vertebrate species.

Lightweight mixture-of-experts medical vision-language model routing visual question answering and image classification to domain-specific experts.

Cell Painting microscopy foundation model, a channel-agnostic masked autoencoder producing morphological embeddings for zero-shot phenotypic analysis.

Framework turning single-cell expression profiles into ranked gene-name sequences, letting off-the-shelf language models generate and annotate cells.

Chinese medical vision-language model pairing a Vision Transformer with an LLM to caption medical images and answer clinical questions in Chinese.

Chest X-ray embedding model built on ELIXR, producing image and image-text embeddings for data-efficient and zero-shot radiograph classification.

Medical image segmentation foundation model for 3D CT and MRI, covering 127 anatomical classes automatically plus interactive point-prompt refinement.

All-atom generative model for de novo protein and peptide binder design against diverse biomolecular targets, wet-lab validated across 26 targets.

Single-cell foundation model with 800M parameters trained on ~100 million human cells, for annotation, perturbation prediction, and gene analysis.

Instruction-tuned vision-language foundation model for chest X-ray interpretation, with 8 billion parameters spanning eight clinical task types.

Chest X-ray vision-language model that drafts the findings section of a radiology report, at 7B parameters small enough to run on a single GPU.

Parameter-efficient protein language model that matches larger models such as ESM-2 on protein prediction tasks using under 10% of the parameters.

Open therapeutics foundation models from Google, built on Gemma-2, for drug-discovery property prediction and conversational reasoning.

Single-cell foundation model pre-trained on 22 million transcriptomes, using rank-based gene encoding for clustering and trajectory inference.

Histopathology foundation model extracting general-purpose features from H&E patches by distilling the UNI, Phikon, and CONCH pathology encoders.

Single-cell foundation model pre-trained on 50 million cells for gene network inference, denoising, and cell type prediction.

Single-cell foundation model pre-trained on 50 million cells that infers cell-specific gene regulatory networks from transformer attention matrices.

Sparse mixture-of-experts autoregressive protein language model family pretrained on 1.5 trillion amino acid tokens with compute-optimal scaling.

Masked DNA language model trained on 800+ species with explicit species conditioning, separating conserved regulatory motifs from background bias.

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

Masked DNA language model trained on over 800 vertebrate genomes and conditioned on species identity to learn conserved regulatory sequence features.

DNA embedding model built on DNABERT-2, using contrastive learning to cluster sequences by species for metagenomic binning without labeled data.

Transformer-based generative language model for de novo RNA design, pretrained on 16 million non-coding RNA sequences from RNAcentral.

ECG foundation model that treats heartbeats as words and rhythm strips as sentences, using heartbeat-level tokenization for diagnostic classification.

Respiratory acoustic foundation models pretrained on roughly 136K cough and breathing recordings for disease detection and lung function estimation.

Grounded multimodal language model for endoscopic surgery, supporting visual dialogue, region-based question answering, and bounding-box grounding.

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

Multimodal LLM unifying 12-lead ECG time series, ECG images, and text for grounded, clinician-aligned electrocardiogram interpretation.

Biomedical multimodal LLM that answers questions about medical images and returns pixel-level segmentation masks, using a mixture-of-experts design.

Multimodal clinical foundation model reasoning jointly over 2D and 3D medical images, ECG time-series, and text reports across nine clinical domains.

Masked-autoencoder foundation model that pre-trains a 3D Residual Encoder U-Net on roughly 39,000 brain MRIs for volumetric image segmentation.

Chest X-ray vision-language model that generates free-text radiology reports, pairing a CXR-specific image encoder with a 7B LLaMA-2 language model.

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

Unified science foundation model treating molecules, proteins, RNA, DNA, and materials as one sequence language, in 1B, 8B, and 46.7B sizes.

Health acoustics foundation model that turns short clips of coughs and breaths into embeddings for building acoustic biomarker models with less data.

Convolutional ECG foundation model trained on expert annotations spanning 150 diagnostic categories, with 12-lead and single-lead wearable variants.

Multimodal foundation model for wearable physiological sensing across PPG, ECG, EEG, GSR, and IMU signals, using channel-aware attention.

Vision Transformer for cell instance segmentation and classification in H&E whole-slide images, extended by CellViT++ with foundation backbones.

Multimodal large language model that interprets 12-lead electrocardiogram images, answering open-ended clinical questions and generating ECG reports.

De novo protein design from natural language: a 16B-parameter framework turning text descriptions into sequences via structure-conditioned generation.

Region-aware bilingual medical multimodal LLM that handles image- and region-level vision-language tasks across eight imaging modalities.

Generalist medical vision-language foundation model with 40B parameters, spanning radiology, pathology, dermatology, retinography, and endoscopy.

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

RNA foundation model with 1.6 billion parameters, pretrained on 42 million non-coding RNA sequences for structure prediction and RNA sequence design.

Universal brain lesion segmentation for multi-modal brain MRI, using a Mixture of Modality Experts to span diverse modalities and lesion types.

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

Tri-modal protein language model aligning sequence, structure, and text in one embedding space for natural-language search over billions of proteins.

Multimodal 80B-parameter protein-language model that answers natural language questions about protein function from sequence and structure.

Mixture-of-experts protein language model scaling to 16 billion parameters, applied to variant effect prediction and de novo protein design.

Flow-matching generative model for de novo atomistic protein binder design against protein and small-molecule targets, including carbohydrate binders.

Vision-language foundation model for fetal ultrasound, pretrained on 210,035 image-text pairs for plane classification, biometry, and segmentation.

Bilingual Arabic-English medical multimodal model built on Llama 3.1 for radiology, CT, and histology image understanding and question answering.

Multi-scale ECG-language model that aligns 12-lead ECG signals with clinical text at token, beat, and rhythm levels for zero-shot cardiac diagnosis.

Single-cell RNA-seq foundation model pretrained on 50 million human cells, encoding the full transcriptome for annotation and perturbation modeling.

Mamba-based mature RNA foundation model, contrastively trained on splice isoforms and 400+ mammalian species orthologs for mRNA property prediction.

DNA foundation model scaling an encoder-only transformer to 7 billion parameters for variant effect prediction, gene expression, and sequence design.

EEG foundation model whose learned queries map any electrode montage into a fixed latent space, scaling linearly in the number of channels.

Perturbation-trained single-cell foundation models (up to 3B parameters) that jointly model genes, cells, and compounds for precision oncology tasks.

ECG foundation model pretrained on 12-lead waveforms paired with clinical reports, enabling label-efficient and zero-shot cardiac diagnosis.

Medical multimodal LLM (2B and 8B) trained for generalizable, step-by-step clinical reasoning via Mentor-Intern Collaborative Search.

LoRA and QLoRA fine-tuning of ESM-2 for token-level prediction of protein binding sites and post-translational modification sites from sequence alone.

Latent diffusion model for controllable all-atom protein generation that co-designs sequence and structure while training on sequences alone.

Histopathology foundation model that encodes 224x224 H&E patches into compact 384-dimensional embeddings for tumor and biomarker classifiers.

Single-cell foundation model using tabular attention over context cells to predict responses to arbitrary perturbations without fine-tuning.

Multimodal foundation model integrating protein sequence, structure, and natural language to model and generate protein phenotypes across scales.

Protein language model trained on roughly 2.8 billion sequences, forming the representation core of Biohub's world model of protein biology.

Predicts single-cell scRNA-seq coverage and scATAC-seq insertion profiles from DNA sequence, adapting the Borzoi trunk with a cell-specific decoder.

Structure-prediction and design engine that turns ESMC sequence representations into all-atom 3D structures of proteins and biomolecular complexes.

Retrieval-augmented protein encoders that fuse homologous sequences into a single-pass transformer for variant effect and contact prediction.

Multimodal protein language model extending ESM-2 and SaProt with a Structure Adapter over residue torsion angles for protein function prediction.

Universal foundation model that jointly generates diagnostic text and segments the corresponding targets across ten biomedical imaging modalities.

Multi-language transformer framework using five pre-trained language models to predict DNA methylation (6mA, 4mC, 5hmC) across species.

Biomedical vision-language assistant for medical visual question answering, pairing Phi-2 with a vision encoder in a 4.2B-parameter model.

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

Multi-modal, multi-task biological foundation model trained on 2 billion samples spanning proteins, small molecules, and single-cell gene expression.

Single-cell foundation model learning technology-agnostic cell embeddings by contrasting cell views rather than reconstructing gene expression counts.

Molecular foundation model that late-fuses graph, image, and SMILES encoders into one embedding for molecular property and drug target prediction.

Protein language models trained on biophysical dynamics from MD simulations and normal-mode analysis; ESMDance builds on ESM2 for variant effects.

Diffusion language model with 4.9 billion parameters that predicts genome-wide CRISPRi perturbation responses in single-cell transcriptomes.

Bidirectional transformer for DNA using k-mer tokenization, fine-tunable for promoter, splice site, and transcription factor binding prediction.

Generative foundation model for cryo-EM density maps using flow matching, enabling zero-shot denoising, map sharpening, and missing wedge restoration.

Conversational T5-based framework that turns scRNA-seq data into cell sentences for cell type annotation and drug sensitivity prediction.

Histopathology vision transformer with 1.1B parameters, pretrained on patches from 500,000 H&E whole-slide images across 4,000 clinical practices.

Reasoning-guided foundation model for de novo antibody CDR design, pairing a multimodal LLM understanding expert with a Boltz-1 diffusion expert.

Multimodal viral foundation model over nucleotide and protein sequence, built for virus discovery, function annotation, and antibody design.

Geometric deep learning model generating context-aware protein representations across 156 cell-type contexts from a multi-organ single-cell atlas.

Histopathology foundation model with 1.1B parameters, trained entirely on public data using JEDI, a dual-stage strategy combining JEPA and DINO.

Hierarchical single-cell foundation model that turns scRNA-seq profiles into zero-shot donor-level embeddings for disease and biomarker prediction.

De novo peptide binder design framework that targets specific motifs, including disordered regions and conserved epitopes, from target sequence alone.

Generalist neuroimaging vision foundation model pretrained on 5.24M clinical MRI and CT volumes for radiologic diagnosis and report generation.

Google's dermatology image embedding model that produces 6144-dimensional embeddings for data-efficient skin-condition classifiers.

Protein sequence encoder that maps ESM2 embeddings to a learned 20-letter alphabet for structure-quality remote homology detection at MMseqs2 speed.

Histology vision transformer with 80M parameters that predicts spatial gene expression from H&E tissue images and transfers to tumor detection.

RNA foundation model pretrained jointly on sequences and secondary structures for structure prediction, homology and splice site classification.

RNA inverse-folding model that generates sequences predicted to fold into a target 3D backbone, capturing non-canonical pairs and tertiary motifs.

Multimodal diffusion model that co-designs protein sequence and 3D structure around cofactors and small molecules for de novo heme enzyme design.

Virtual cell model using masked discrete diffusion over the whole transcriptome to simulate scRNA-seq perturbation responses across tissues.

DNA language model for genome-wide variant effect prediction, trained by masked language modeling on multispecies genomes with no labeled data.

fMRI foundation model trained on cortical flat-map videos with masked autoencoding, showing power-law scaling on brain activity reconstruction.

Proteome-scale protein language model whose representations enable zero-shot protein-protein interaction and gene essentiality prediction.

Generative DNA foundation model trained on 91.7M nucleotide sequences and annotations for species classification and mutation effect prediction.

LoRA adapter on ProstT5 predicting per-residue distributions over Foldseek 3Di tokens, capturing conformational flexibility from MD trajectories.

Chemical language models pretrained on SMILES for therapeutic peptides, natively representing non-canonical residues, cyclization, and conjugation.

Single-cell foundation model adapting LLaMA-3.1-8B with LoRA, recasting transcriptomes and protein interaction networks as natural-language Q&A pairs.

Post-hoc method that restores monotonic scaling to ESM-2 embeddings, yielding Matryoshka-style nested representations for variant effect prediction.

Transformer foundation model for single-cell ATAC-seq that embeds both cells and cis-regulatory elements for annotation and batch correction.

Sequence-based discrete-diffusion framework that designs peptide binders with specified agonist or antagonist behavior against GPCR targets.

Pathology foundation model that aligns whole-slide images with genomic, epigenetic, and transcriptomic data for patient-level tumor representations.

Protein structure tokenizer that maps 3D backbones to discrete tokens with an SE(3)-equivariant encoder preserving orientation and chirality.

Protein-family language model trained on unaligned homolog sets for zero-shot variant fitness prediction and design. ProFam-1 holds 251M parameters.

Autoregressive model for therapeutic mRNA design that jointly generates 5' UTR, CDS, and 3' UTR, pretrained on 30 million full-length natural mRNAs.

Contrastive model built on a genomic language model that predicts physical protein-protein interactions across a microbial proteome in linear time.

Molecular linker design model fine-tuned from Llama 3 that emits PROTAC and fragment linkers as SMILES from natural-language geometry prompts.

Small-molecule drug discovery foundation model covering ADMET, retrosynthesis, drug-target activity, and molecular optimization in a 2.6B checkpoint.

Molecular foundation models pretrained on density functional theory data, encoding 3D geometry and quantum behavior for ADMET and drug discovery.

NMR foundation model trained on 158 million simulated 1H and 13C spectra, transferring simulation-learned representations to real experimental data.

Generative RNA foundation model trained on 114 million full-length sequences for de novo design of tRNAs, aptamers, CRISPR guide RNAs, and mRNAs.

Conditional codon language model with 150M parameters that generates species-optimized coding sequences from a protein and its taxonomic lineage.

Pan-viral genomic language model producing fixed genome-level embeddings of viral DNA and RNA, reused across classification tasks without retraining.

SE(3)-invariant masked autoencoder that learns protein fold representations from AlphaFold-DB structures, supporting zero-shot fold classification.

DNA foundation model using masked discrete diffusion to unify bidirectional sequence understanding and de novo generation in one architecture.

Self-supervised Siamese network for cryo-electron tomography, enabling zero-shot denoising, segmentation, and macromolecule detection in tomograms.

Family of autoregressive genomic foundation models that reconcile k-mer tokenization with single-nucleotide resolution at contexts up to 98k bp.

Plant genomic foundation models from 0.1B to 1B parameters, pretrained on 43 phylogenetically diverse plant genomes for variant effect prediction.

Self-supervised SE(3) geometric pretraining for protein backbone generators, improving designability, motif scaffolding, and conformational ensembles.

110M-parameter RNA language model that designs sequences from secondary structure, motif, and Gene Ontology constraints via discrete diffusion.

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

Transformer that classifies tumour types and subtypes from somatic variants in whole-genome and whole-exome data, with auto-downloading checkpoints.

Genomic foundation model for Cypriniformes fish, built on a Mamba-2 state space model with a 32 kb context window for long-range genome modeling.

Cross-species multimodal foundation model of immunology and inflammation, harmonizing transcriptomics and histology into patient-level embeddings.

Universal all-atom machine-learning force field for molecular dynamics, with ab initio-level accuracy on solvated biomolecules of ~1,500 atoms.

Cryo-EM ligand modeling pipeline that detects bound ligand densities in a map, then reconstructs their atomic structures with a diffusion model.

Diffusion transformer for virtual tissue synthesis, generating H&E histopathology patches conditioned on spatial gene expression and morphology.

Promptable DNA language model that generates multi-kilobase plasmid sequences from plain-language component specs, refined with verifiable rewards.

Bidirectional state-space (Mamba-2) genomic model for ultra-long extrachromosomal circular DNA, scaling linearly with sequence length.

Family of ten compact GPT-2 decoder-only DNA language models spanning BPE vocabularies from 16 to 8192 tokens, built for lossless genome compression.

Masked language model for T-cell receptor and peptide-MHC binding prediction, with compositional pretraining and non-autoregressive decoding.

Plant genome foundation model pairing a bidirectional Mamba backbone with sparse Mixture-of-Experts, pretrained on 25.4B nucleotides from 42 species.

Cryo-EM foundation model pre-trained on 65 million particle images, enabling zero-shot classification, pose clustering, and quality assessment.

Generative model for chemically modified and macrocyclic peptides that builds molecules in HELM notation, supporting de novo design and infilling.

Generative single-cell foundation model trained on 100M+ transcriptomes that predicts how genetic perturbations reshape cell trajectories over time.

Coding-sequence foundation model for mRNA design, pretrained as a BART denoising encoder-decoder on mRNA from nine taxonomic groups.

Protein sequence embedding model, contrastively fine-tuned from ESM-2, that places functionally and structurally related proteins close together.

Causal 309M-parameter protein language model that scores variant fitness zero-shot and generates sequences, reaching 0.390 Spearman on ProteinGym.

Autoregressive DNA foundation model for variant effect prediction, using 6-mer tokenization to match Evo2-7B win rates at far higher throughput.

Virtual cell foundation model pretrained on over 23 million cells from 5,000 patient samples for drug target and biomarker discovery.

Peptide language model trained on HELM notation, a DeBERTa encoder for property prediction on macrocyclic and non-canonical medium-sized peptides.

Whole-cell segmentation model for spatial transcriptomics that fuses DAPI nuclear images with RNA transcript density to recover true cell boundaries.

Predicts protein complex stoichiometry from amino acid sequence alone, ranking copy numbers in seconds and exporting AlphaFold3-ready JSON files.

Three fixed ProtGPT2 fine-tunes specialized for metalloprotein generation, trained on ProteinMPNN-derived synthetic sequences.

Graph transformer foundation model for glycans, learning reusable embeddings of branched carbohydrate structures for glycomics prediction tasks.

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

Retrieval-augmented genomic foundation model that gives transformer backbones a hash-based k-mer motif memory for functional genomics tasks.

Genetically aligned foundation model for blood smear cytology that links single-cell morphology to the chromosomal aberrations behind AML and APL.

Contrastive promoter-protein pretraining that aligns bacterial promoters with their encoded proteins to learn regulatory genomics representations.

Bimodal masked language model that jointly encodes bulk RNA-seq expression and DNA methylation into patient-level embeddings for cancer genomics.

SMILES molecular encoder on a DeBERTaV2 backbone, pretrained on 123M PubChem molecules with physicochemical and structural-similarity objectives.

Multimodal molecular generation model for drug design, conditioned on properties, pharmacophores, protein sequences, or protein binding pockets.

Plant genome foundation model for ab initio gene structure annotation, predicting genes, coding sequences, and exons at single-nucleotide resolution.

470M-parameter microbial genome foundation model trained on 234.5B base pairs for multi-scale genomic representation and trait prediction.

Hyperbolic protein language model for alignment-free phylogenetic inference, turning ESM2-650M embeddings into distance matrices for tree placement.

Histopathology model that predicts single-cell type composition and reconstructs spatial gene expression from H&E slides, with no molecular assay.

Antibody language model pretrained only on CDR-H3 loops, giving embeddings for immune repertoire analysis and antibody sequence classification.

Mixture-of-Experts genomic foundation model for the human microbiome, with 4.7B parameters pretrained on bacterial, archaeal, and phage genomes.

Microbiome world model that treats a community as a set of taxa, scoring how well each member fits and predicting community dynamics zero-shot.

Long-context plant DNA language model, 676M parameters on a Mamba2 backbone, pretrained on 65 angiosperm genomes for cross-species variant annotation.

Generative microscopy foundation model that synthesizes in-silico fluorescence images of protein subcellular localization from amino-acid sequence.

Multi-target drug discovery framework pairing a diffusion-transformer generator with evolutionary latent-space search and synthesis-aware scoring.

Vision-language foundation model pre-trained on screening mammogram-report pairs to improve data efficiency and robustness in breast cancer detection.

Decoder-only foundation model that unifies sequences, 3D structures, and natural language for small molecules and proteins in one shared token space.

Genomic foundation model for rice, pretrained on 422 Oryza genomes with a 1 Mbp context window and a 1.25B-parameter mixture-of-experts transformer.

Open-source framework for building RNA and DNA foundation models, featuring WCED pretraining for transcriptomics and SNP-aware encoding for genomics.

Text-to-text biological language model spanning molecules, proteins, and text, adding IUPAC names and multi-task instruction tuning to BioT5.

Encoder-decoder framework unifying molecules, proteins, and natural language with SELFIES notation for cross-modal drug discovery tasks.

Single-cell latent diffusion model fine-tuned on 14.5 million CD4+ T cells to simulate transcriptomic effects of single-gene perturbations.

Protein language model family at 300M, 600M, and 6B parameters, purpose-built for representation learning and outperforming ESM-2 at smaller scale.

Codon-resolution language models trained on 130 million coding sequences from 20,000 species, learning codon rules for translation and mRNA stability.

Partially latent flow-matching model for de novo protein design, jointly generating sequence and all-atom structure for proteins up to 800 residues.

Bilingual protein language model that translates bidirectionally between amino acid sequences and the 3Di structural alphabet for inverse folding.

Transformer protein language model trained on 250 million protein sequences that learns structural and functional representations without supervision.

Family of transformer-based DNA language models using BPE tokenization and BigBird sparse attention to reach context lengths up to 36,000 base pairs.