# Biology — Hugging Science

> Life sciences, genomics, and biological systems

_Auto-generated from site data. Fetch `/llms.txt` for the full index._

## Datasets (104)

### arcinstitute/opengenome2
- **Type**: Genomics
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/datasets/arcinstitute/opengenome2

Curated collection of prokaryotic and eukaryotic genomic sequences for training and benchmarking large-scale biological foundation models.

### arcinstitute/SE-167M-Human
- **Type**: Single-Cell Biology
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/datasets/arcinstitute/SE-167M-Human

167M human single-cell RNA expression profiles across diverse tissues and cell types, used for training STACK and SE single-cell foundation models.

### arcinstitute/Stack-CellxGene45M
- **Type**: Single-Cell Biology
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/datasets/arcinstitute/Stack-CellxGene45M

45M curated single-cell profiles drawn from the CellxGene corpus, standardised for in-context learning and cross-study perturbation analysis.

### ginkgo-datapoints/GDPa1
- **Type**: Antibody Developability
- **Tags**: Biology, Biotechnology
- **HuggingFace**: https://huggingface.co/datasets/ginkgo-datapoints/GDPa1

Antibody developability dataset with biophysical assay data for 242 antibodies across 9 assays.

### ginkgo-datapoints/GDPx1
- **Type**: Functional Genomics
- **Tags**: Biology, Biotechnology
- **HuggingFace**: https://huggingface.co/datasets/ginkgo-datapoints/GDPx1

DRUG-seq functional genomics dataset with chemical perturbation experiments in A549 cells.

### ginkgo-datapoints/GDPx2
- **Type**: Functional Genomics
- **Tags**: Biology, Biotechnology
- **HuggingFace**: https://huggingface.co/datasets/ginkgo-datapoints/GDPx2

DRUG-seq transcriptomic profiling across 4 primary human cell types with 85 compounds.

### ginkgo-datapoints/GDPx3
- **Type**: Cell Imaging
- **Tags**: Biology, Biotechnology
- **HuggingFace**: https://huggingface.co/datasets/ginkgo-datapoints/GDPx3

High-content Cell Painting imaging dataset for AI/ML model training in drug discovery.

### ginkgo-datapoints/GDPx4
- **Type**: Functional Genomics
- **Tags**: Biology, Biotechnology
- **HuggingFace**: https://huggingface.co/datasets/ginkgo-datapoints/GDPx4

DRUG-seq transcriptomic profiling in engineered HEK293 cells with inducible gene overexpression, enabling systematic study of gene-drug interactions.

### eve-bio/drug-target-activity
- **Type**: Drug Discovery
- **Tags**: Biology, Medicine, Chemistry
- **HuggingFace**: https://huggingface.co/datasets/eve-bio/drug-target-activity

Drug-target interaction measurements for 1,397 FDA-approved small molecule drugs.

### EarthSpeciesProject/BEANS-Zero
- **Type**: Bioacoustics Benchmark
- **Tags**: Biology, Ecology, Conservation, Benchmark, Earth Science
- **HuggingFace**: https://huggingface.co/datasets/EarthSpeciesProject/BEANS-Zero

Zero-shot bioacoustics benchmark evaluating audio-language models on species detection, classification, and captioning across diverse animal taxa.

### SandboxAQ/SAIR
- **Type**: Drug Discovery
- **Tags**: Chemistry, Medicine, Biology
- **HuggingFace**: https://huggingface.co/datasets/SandboxAQ/SAIR

Largest public dataset of protein-ligand 3D structures with binding affinity measurements (1M+ pairs).

### google/spiqa
- **Type**: Scientific Benchmark
- **Tags**: Biology, Chemistry, Physics, Benchmark, Scientific Reasoning, Mathematics
- **HuggingFace**: https://huggingface.co/datasets/google/spiqa

Scientific Paper Image Question Answering benchmark requiring multimodal reasoning over figures, charts, and diagrams from research papers across scientific domains.

### InstaDeepAI/NTv3_benchmark_dataset
- **Type**: Genomics
- **Tags**: Biology, Genomics, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/InstaDeepAI/NTv3_benchmark_dataset

Benchmark dataset with functional tracks and genome annotations across 7 species.

### InstaDeepAI/nucleotide_transformer_downstream_tasks
- **Type**: Genomics
- **Tags**: Biology, Genomics, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/InstaDeepAI/nucleotide_transformer_downstream_tasks

18 genomic prediction benchmark tasks covering histone marks, regulatory regions, splice sites, and promoter activity across human and multi-species genomes.

### InstaDeepAI/multi_species_genomes
- **Type**: Genomics
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/InstaDeepAI/multi_species_genomes

Whole-genome sequences for 850 species spanning bacteria, fungi, plants, and animals — the pre-training corpus for the Nucleotide Transformer model family.

### InstaDeepAI/plant-genomic-benchmark
- **Type**: Plant Genomics
- **Tags**: Biology, Genomics, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/InstaDeepAI/plant-genomic-benchmark

Plant genomics benchmark spanning gene expression, chromatin accessibility, and agronomic trait prediction tasks across multiple crop and model plant species.

### InstaDeepAI/winnow-ms-datasets
- **Type**: Proteomics
- **Tags**: Biology, Chemistry
- **HuggingFace**: https://huggingface.co/datasets/InstaDeepAI/winnow-ms-datasets

Mass spectrometry datasets for protein analysis and ML model training.

### InstaDeepAI/true-cds-protein-tasks
- **Type**: Protein Tasks
- **Tags**: Biology, Genomics, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/InstaDeepAI/true-cds-protein-tasks

Coding sequence and protein function prediction benchmark tasks.

### futurehouse/BixBench
- **Type**: Research Benchmark
- **Tags**: Biology, Chemistry, Benchmark, Scientific Reasoning, Mathematics
- **HuggingFace**: https://huggingface.co/datasets/futurehouse/BixBench

Benchmark with 205 reproducible research questions paired with data capsules for AI evaluation.

### futurehouse/lab-bench
- **Type**: Research Benchmark
- **Tags**: Biology, Benchmark, Scientific Reasoning, Mathematics
- **HuggingFace**: https://huggingface.co/datasets/futurehouse/lab-bench

Language Agent Biology Benchmark - 8 categories of scientific research tasks including cloning, figures, and protocols.

### tahoebio/Tahoe-100M
- **Type**: Single-Cell Biology
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/datasets/tahoebio/Tahoe-100M

Giga-scale perturbation atlas with 100M+ single-cell profiles from 50 cancer cell lines and 1,100 drugs.

### tahoebio/Tahoe-x1-embeddings
- **Type**: Single-Cell Biology
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/datasets/tahoebio/Tahoe-x1-embeddings

Pre-computed cell and gene embeddings from the Tahoe-x1 foundation model.

### tahoebio/EmeraldBay
- **Type**: Single-Cell Perturbation
- **Tags**: Biology, Medicine, Chemistry, Genomics
- **HuggingFace**: https://huggingface.co/datasets/tahoebio/EmeraldBay

Emerald Bay — single-cell perturbation dataset of 1.8M+ transcriptomic profiles spanning 52 cell lines × 91 drug treatments (plus combinations), generated on Tahoe’s MOSAIC high-throughput platform with paired transcriptional and drug-phenotype readouts over a five-day culture.

### tahoebio/replogle-nadig-de-rhaister
- **Type**: CRISPR Differential Expression
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/datasets/tahoebio/replogle-nadig-de-rhaister

Differential-expression summary statistics for the Replogle-Nadig genome-scale CRISPR screen (4 cell lines × ~2,000 gene knockdowns: HepG2, Jurkat, K562, RPE1) — per-gene fold changes, significance, and pseudobulk deltas used to train Rhaister.

### tahoebio/tahoe-de-rhaister
- **Type**: Drug Differential Expression
- **Tags**: Biology, Medicine, Chemistry
- **HuggingFace**: https://huggingface.co/datasets/tahoebio/tahoe-de-rhaister

Differential-expression summary statistics for the Tahoe-100M drug screen — 50 cancer cell lines × 384 drugs × 3 doses summarized from ~100M raw single cells, including per-gene fold changes, significance, and pseudobulk deltas (pdex + cell_eval tables) used to train Rhaister.

### tahoebio/parse-de-rhaister
- **Type**: Cytokine Differential Expression
- **Tags**: Biology, Medicine
- **HuggingFace**: https://huggingface.co/datasets/tahoebio/parse-de-rhaister

Differential-expression summary statistics for the Parse PBMC cytokine screen (12 donors × 18 immune cell types × 90 cytokines), derived from the Arc Institute ST-HVG-Parse dataset — per-gene fold changes, significance, and pseudobulk deltas used to train Rhaister.

### MahmoodLab/hest
- **Type**: Spatial Transcriptomics
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/datasets/MahmoodLab/hest

HEST-1k — 1,276 spatial-transcriptomic profiles each linked and aligned to a Whole Slide Image (pixel size <1.15 µm/px), the largest paired histology + spatial-transcriptomics resource on the Hub.

### MahmoodLab/hest-bench
- **Type**: Pathology Benchmark
- **Tags**: Medicine, Biology, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/MahmoodLab/hest-bench

HEST-Bench — companion benchmark to HEST-1k for evaluating spatial-transcriptomics and pathology foundation models on aligned histology / expression tasks.

### MahmoodLab/UNI2-h-features
- **Type**: Pathology Features
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/datasets/MahmoodLab/UNI2-h-features

Precomputed UNI2-h patch features for TCGA, CPTAC, and PANDA at 256×256 / 20× — drop-in feature set for downstream pathology models without recomputing the UNI2-h forward pass.

### MahmoodLab/Patho-Bench
- **Type**: Pathology Benchmark
- **Tags**: Medicine, Biology, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/MahmoodLab/Patho-Bench

Patho-Bench — benchmark designed to evaluate patch and slide encoder foundation models for whole-slide images across cancer subtyping, biomarker prediction, and survival tasks.

### MahmoodLab/TCGA-UniformTumor-8K
- **Type**: Pathology Benchmark
- **Tags**: Medicine, Biology, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/MahmoodLab/TCGA-UniformTumor-8K

TCGA-UniformTumor-8K — region-level pan-cancer subtyping resource with 25,495 ROIs of 8,192×8,192 px tiled from TCGA whole-slide images, designed for high-resolution slide-encoder evaluation.

### MahmoodLab/viper
- **Type**: Pathology VLM Benchmark
- **Tags**: Medicine, Biology, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/MahmoodLab/viper

VIPER — expert-curated benchmark of 1,251 questions over 419 H&E-stained rat histology images for evaluating vision-language models in veterinary and toxicologic pathology.

### owkin/plism-dataset-tiles
- **Type**: Computational Pathology
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/datasets/owkin/plism-dataset-tiles

Large-scale histopathology tile dataset for benchmarking robustness of pathology foundation models across staining and scanner variability.

### owkin/nct-crc-he
- **Type**: Computational Pathology
- **Tags**: Medicine, Biology, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/owkin/nct-crc-he

Colorectal cancer tissue classification dataset with H&E-stained patches across 9 tissue classes, widely used for benchmarking pathology models.

### owkin/camelyon16-features
- **Type**: Computational Pathology
- **Tags**: Medicine, Biology, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/owkin/camelyon16-features

Pre-extracted features from the CAMELYON16 breast cancer lymph node metastasis detection challenge, enabling efficient benchmarking of MIL methods.

### owkin/her2-challenge-2026
- **Type**: Computational Pathology
- **Tags**: Medicine, Biology, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/owkin/her2-challenge-2026

HER2 scoring challenge dataset with H&E-stained whole-slide images for evaluating AI-based HER2 status prediction in breast cancer.

### Xaira-Therapeutics/X-Atlas-Orion
- **Type**: Single-Cell Biology
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/datasets/Xaira-Therapeutics/X-Atlas-Orion

Large-scale single-cell transcriptomics atlas with millions of cell profiles from diverse human tissues, designed for training perturbation-aware foundation models.

### Xaira-Therapeutics/X-Atlas-Pisces
- **Type**: Single-Cell Biology
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/datasets/Xaira-Therapeutics/X-Atlas-Pisces

CRISPRi perturbation single-cell dataset pairing genetic knockdowns with transcriptomic responses, used for training and evaluating the X-Cell model.

### AllTheBacteria/ATB
- **Type**: Microbial Genomics
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/AllTheBacteria/ATB

AllTheBacteria: a comprehensive collection of ~2 million bacterial genome assemblies from public sequence databases, standardized for large-scale genomic analysis.

### AllTheBacteria/Bac-Corpus-protein-sequences-high-diversity
- **Type**: Microbial Genomics
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/AllTheBacteria/Bac-Corpus-protein-sequences-high-diversity

High-diversity corpus of bacterial protein sequences derived from the ATB collection, filtered for maximum sequence diversity to support protein language model pretraining.

### AllTheBacteria/Bac-Corpus-dna-intergenic-sequences-high-diversity
- **Type**: Microbial Genomics
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/AllTheBacteria/Bac-Corpus-dna-intergenic-sequences-high-diversity

High-diversity corpus of bacterial intergenic DNA sequences for training DNA language models on non-coding regulatory regions.

### AllTheBacteria/SPIRE
- **Type**: Microbial Genomics
- **Tags**: Biology, Genomics, Ecology, Earth Science
- **HuggingFace**: https://huggingface.co/datasets/AllTheBacteria/SPIRE

Searchable Planetary-scale mIcrobiome REsource: a large-scale metagenomics resource aggregating environmental microbiome samples from diverse global habitats.

### opig/OAS
- **Type**: Antibody Sequences
- **Tags**: Biology, Medicine, Chemistry
- **HuggingFace**: https://huggingface.co/datasets/opig/OAS

Observed Antibody Space: a curated database of over one billion antibody sequences from immune repertoire sequencing studies, the standard resource for antibody ML.

### wanglab/CT_DeepLesion-MedSAM2
- **Type**: Medical Imaging
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/datasets/wanglab/CT_DeepLesion-MedSAM2

CT volumes from the DeepLesion benchmark with mask annotations restructured for training and evaluating MedSAM2, the universal medical image segmentation foundation model.

### wanglab/img_virus_plasmid
- **Type**: Microbial Genomics
- **Tags**: Biology, Genomics, Biotechnology
- **HuggingFace**: https://huggingface.co/datasets/wanglab/img_virus_plasmid

Combined IMG/VR (uncultivated virus genomes) and IMG/PR (plasmids from genomes and metagenomes) catalog with rich functional, taxonomic, and ecological metadata.

### wanglab/kegg
- **Type**: Biological Reasoning
- **Tags**: Biology, Genomics, Scientific Reasoning
- **HuggingFace**: https://huggingface.co/datasets/wanglab/kegg

KEGG pathway entries paired with variant annotations for training and evaluating multimodal biological reasoning models (used by the BioReason work).

### OpenMed/synthvision-annotated-qwen
- **Type**: Synthetic Medical Vision
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/datasets/OpenMed/synthvision-annotated-qwen

Synthetic medical-imaging dataset annotated by Qwen — used in OpenMed’s SynthVision pipeline for training and validating medical multimodal models.

### OpenMed/synthvision-seeds
- **Type**: Synthetic Medical Vision
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/datasets/OpenMed/synthvision-seeds

Seed prompts and source imagery feeding the SynthVision generation pipeline that produces OpenMed’s annotated medical-imaging training corpora.

### OpenMed/synthvision-annotated-kimi
- **Type**: Synthetic Medical Vision
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/datasets/OpenMed/synthvision-annotated-kimi

Synthetic medical-imaging dataset annotated by Kimi — sister set to the Qwen-annotated split, supporting cross-annotator validation in the SynthVision pipeline.

### allenai/peS2o
- **Type**: Pretraining Corpus
- **Tags**: Scientific Reasoning, Biology, Chemistry, Physics, Engineering
- **HuggingFace**: https://huggingface.co/datasets/allenai/peS2o

Approximately 40M cleaned, filtered, and formatted open-access academic papers derived from S2ORC — a large multi-domain pretraining corpus for science-aware language models, spanning biology, chemistry, engineering, computer science, and physics.

### Anthropic/BioMysteryBench-preview
- **Type**: Biology Benchmark
- **Tags**: Biology, Medicine, Scientific Reasoning, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/Anthropic/BioMysteryBench-preview

Preview slice of BioMysteryBench — challenging, expert-curated biology problems for evaluating AI scientific reasoning capability.

### Anthropic/BioMysteryBench-full
- **Type**: Biology Benchmark
- **Tags**: Biology, Medicine, Scientific Reasoning, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/Anthropic/BioMysteryBench-full

Full BioMysteryBench evaluation set — challenging biology problems used to probe expert-level scientific reasoning in frontier models.

### maomlab/Molecule3D
- **Type**: Molecular Properties
- **Tags**: Chemistry, Biology
- **HuggingFace**: https://huggingface.co/datasets/maomlab/Molecule3D

Curated 3D molecular structures with computed properties — supports geometric deep learning for property prediction and conformer-aware modelling.

### maomlab/TDC
- **Type**: Therapeutics Benchmark
- **Tags**: Medicine, Chemistry, Biology
- **HuggingFace**: https://huggingface.co/datasets/maomlab/TDC

Therapeutics Data Commons subset — drug-discovery tasks (ADMET, drug-target interaction, generation) curated for benchmarking molecular ML.

### maomlab/CryptoCEN
- **Type**: Coexpression Network
- **Tags**: Biology, Medicine
- **HuggingFace**: https://huggingface.co/datasets/maomlab/CryptoCEN

CryptoCEN — Cryptococcus coexpression network dataset for fungal pathogen biology and drug-target prioritisation.

### imageomics/TreeOfLife-200M
- **Type**: Biodiversity Image Corpus
- **Tags**: Biology, Ecology, Conservation
- **HuggingFace**: https://huggingface.co/datasets/imageomics/TreeOfLife-200M

Foundational 200M-image dataset for organismal biology — multilingual species labels (en, la) at biodiversity scale, used to train BioCLIP-2 for zero-shot species classification.

### Aignostics/OpenTME
- **Type**: Digital Pathology
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/datasets/Aignostics/OpenTME

Pre-analyzed H&E whole-slide images from TCGA across breast, bladder, colorectal, liver, and lung cancers — cell-level annotations and tumour-microenvironment spatial features generated by Atlas H&E-TME.

### wanglab/bioreason-pro-sft-reasoning-data
- **Type**: Biological Reasoning Corpus
- **Tags**: Biology, Genomics, Scientific Reasoning
- **HuggingFace**: https://huggingface.co/datasets/wanglab/bioreason-pro-sft-reasoning-data

Reasoning trace dataset used to supervised-fine-tune BioReason-Pro — multimodal biological problems with rationales over genomic variants and pathway data.

### HuggingFaceBio/perturbation-bench
- **Type**: Genomic Benchmark
- **Tags**: Biology, Genomics, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/HuggingFaceBio/perturbation-bench

Genetic perturbation benchmark — evaluates how well genomic foundation models predict perturbation responses across cell states.

### HuggingFaceBio/clinvar-vep
- **Type**: Clinical Variants
- **Tags**: Biology, Genomics, Medicine, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/HuggingFaceBio/clinvar-vep

ClinVar variants annotated with VEP (Variant Effect Predictor) — formatted for benchmarking genomic-LLM pathogenicity classification.

### HuggingFaceBio/traitgym
- **Type**: Trait Prediction Benchmark
- **Tags**: Biology, Genomics, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/HuggingFaceBio/traitgym

TraitGym — multi-task benchmark for evaluating genomic foundation models on trait prediction across diverse phenotypes.

### HuggingFaceBio/genomic-niah
- **Type**: Long-Context Benchmark
- **Tags**: Biology, Genomics, Benchmark, Scientific Reasoning
- **HuggingFace**: https://huggingface.co/datasets/HuggingFaceBio/genomic-niah

Genomic Needle-in-a-Haystack — long-context evaluation for genomic LLMs, measuring retrieval and reasoning over long DNA sequences.

### HuggingFaceBio/carbon-pretraining-corpus
- **Type**: Genomic Pretraining Corpus
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/HuggingFaceBio/carbon-pretraining-corpus

Carbon pretraining corpus — the curated DNA sequence dataset used to train the Carbon-500M / 3B / 8B genomic foundation models.

### DSIMB/PATHOS-PLM-EMBEDDINGS
- **Type**: Protein Embeddings
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/datasets/DSIMB/PATHOS-PLM-EMBEDDINGS

PATHOS protein language-model embeddings — precomputed feature representations supporting pathogenicity prediction and downstream macromolecular ML on protein sequences.

### tattabio/OMG
- **Type**: Genomic Corpus
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/tattabio/OMG

Open Mixed Genomes (OMG) — large mixed-organism nucleotide corpus underpinning Tatta Bio’s gLM2 genomic foundation models.

### tattabio/OG
- **Type**: Genomic Corpus
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/tattabio/OG

Open Genomes (OG) — curated genome-sequence corpus from Tatta Bio for genomic ML pretraining and benchmarking.

### recursionpharma/rxrx3
- **Type**: Phenomics Imaging
- **Tags**: Biology, Medicine, Chemistry
- **HuggingFace**: https://huggingface.co/datasets/recursionpharma/rxrx3

Full RxRx3 release — multi-million image high-content microscopy dataset spanning genetic and chemical perturbations across human cell lines, paired with rich text annotations for image-based drug discovery.

### recursionpharma/rxrx3-core
- **Type**: Phenomics Imaging
- **Tags**: Biology, Medicine, Chemistry
- **HuggingFace**: https://huggingface.co/datasets/recursionpharma/rxrx3-core

Curated core subset of RxRx3 — high-quality phenomics images for benchmarking and lower-cost training of phenomic foundation models like OpenPhenom.

### arcinstitute/Perturb-Sapiens
- **Type**: Single-Cell Perturbation
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/datasets/arcinstitute/Perturb-Sapiens

Large-scale human single-cell perturbation dataset used in the STACK foundation-model lineage — paired baseline and perturbed expression profiles for genetic perturbation screens.

### arcinstitute/Replogle-Nadig-Preprint
- **Type**: Single-Cell Perturbation
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/datasets/arcinstitute/Replogle-Nadig-Preprint

Replogle-Nadig single-cell perturbation dataset (preprint release) — Perturb-seq screens used in the STATE single-cell embedding work for perturbation-response modelling.

### arcinstitute/State-Tahoe-Filtered
- **Type**: Single-Cell Perturbation
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/datasets/arcinstitute/State-Tahoe-Filtered

Filtered Tahoe-100M slice used in the STATE workflow — high-quality single-cell perturbation profiles for training and benchmarking cross-study cell-state models.

### biohub/popsicle
- **Type**: CryoET Benchmark
- **Tags**: Biology, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/biohub/popsicle

POPSICLE (Particle/Object Picking & Segmentation In CryoET Learning Evaluation) — multi-modal cryo-electron tomography benchmark for particle picking, localization, and segmentation across phantom, yeast, bacteria, and motor datasets.

### biohub/DynaCLR-data
- **Type**: Microscopy
- **Tags**: Biology, Medicine
- **HuggingFace**: https://huggingface.co/datasets/biohub/DynaCLR-data

Training and evaluation data for DynaCLR — dynamic contrastive learning of cell embeddings from live-cell microscopy time series, including infection-status labels.

### biohub/ESMC-SAE-Features
- **Type**: Interpretability
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/biohub/ESMC-SAE-Features

Feature table for the 16,384-feature sparse autoencoder trained on layer 60 of ESMC-6B — supports mechanistic-interpretability research on protein language models.

### LiteFold/PDB
- **Type**: Protein Structure
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/PDB

Hub-packaged index over the Protein Data Bank mmCIF entries — the global archive of experimentally-determined 3D structures of biological macromolecules.

### LiteFold/STRING
- **Type**: Protein Interactions
- **Tags**: Biology, Medicine
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/STRING

STRING v12.0 — protein association network integrating experimental, computational, text-mined, and curated evidence for functional and physical protein-protein interactions.

### LiteFold/NCBI
- **Type**: Protein Sequences
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/NCBI

NCBI RefSeq protein shard index — curated non-redundant protein subset of NCBI’s Reference Sequence collection, packaged as parquet for efficient loading.

### LiteFold/UniProtKB
- **Type**: Protein Sequences
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/UniProtKB

Processed UniProtKB release — comprehensive, high-quality protein sequence and functional annotation knowledgebase (Swiss-Prot + TrEMBL) reformatted for ML pipelines.

### LiteFold/Mgnify
- **Type**: Metagenomics
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/Mgnify

MGnify protein catalogues — microbiome-derived protein sequences from EBI’s MGnify platform with train/test splits for downstream metagenomic protein tasks.

### LiteFold/ProteinGym
- **Type**: Protein Benchmark
- **Tags**: Biology, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/ProteinGym

Hub-packaged ProteinGym — the standard benchmark suite for evaluating protein fitness and variant-effect predictors, from the OATML / Marks lab (NeurIPS 2023).

### LiteFold/MegaScale-Tsuboyama2023
- **Type**: Protein Stability
- **Tags**: Biology, Medicine
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/MegaScale-Tsuboyama2023

MegaScale (Tsuboyama et al., 2023) — large-scale experimental measurements of protein stability across hundreds of thousands of mutations for stability prediction modelling.

### LiteFold/AlphaFoldDB
- **Type**: Predicted Structures
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/AlphaFoldDB

AlphaFoldDB prediction index — open database of predicted protein 3D structures with confidence scores, providing structural coverage for known protein sequences at scale.

### LiteFold/OpenProteinSet
- **Type**: MSA / Training Corpus
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/OpenProteinSet

OpenProteinSet — open-source MSA training corpus released by the OpenFold team (Ahdritz et al., NeurIPS 2023) reproducing AlphaFold-style training data.

### LiteFold/UniRef90
- **Type**: Protein Sequences
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/UniRef90

UniRef90 shard index — clustered UniProt sequences at 90% identity, the standard sequence database for protein-LM pretraining and homology search.

### LiteFold/FLIP2
- **Type**: Protein Benchmark
- **Tags**: Biology, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/FLIP2

FLIP2 — second-generation Fitness Landscape Inference for Proteins benchmark (Feb 2026 bioRxiv) spanning NVIDIA, Microsoft, Caltech, and Profluent for protein-engineering eval.

### LiteFold/PDB-CCD
- **Type**: Chemical Components
- **Tags**: Biology, Chemistry
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/PDB-CCD

PDB Chemical Component Dictionary — residues, modified residues, ligands, and solvents used by the Protein Data Bank, parquet-packaged for structural-ML pipelines.

### LiteFold/ESMAtlas
- **Type**: Predicted Structures
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/ESMAtlas

Hub mirror of the ESM Metagenomic Atlas — large-scale predicted metagenomic protein structures generated with ESMFold, packaged as webdataset shards.

### LiteFold/SwissSidechain
- **Type**: Non-Canonical Residues
- **Tags**: Biology, Chemistry
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/SwissSidechain

SwissSideChain — molecular and structural database of non-natural amino-acid sidechains with rotamers and molecular-mechanics parameters for peptide/biologic design.

### LiteFold/UniRef50
- **Type**: Protein Sequences
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/UniRef50

UniRef50 — UniProt reference cluster representatives at 50% identity, the most-aggressively clustered tier widely used for protein-LM pretraining.

### LiteFold/SKEMPI2
- **Type**: Binding Affinity
- **Tags**: Biology, Benchmark
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/SKEMPI2

SKEMPI v2 — manually curated benchmark of experimental changes in protein-protein binding affinity, kinetics, and thermodynamics upon mutation.

### LiteFold/CATH
- **Type**: Domain Classification
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/CATH

CATH domain classification — hierarchical classification of protein domain structures by class, architecture, topology, and homologous superfamily.

### LiteFold/DisProt
- **Type**: Protein Annotation
- **Tags**: Biology, Medicine
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/DisProt

DisProt — manually curated database of intrinsically disordered proteins and regions with experimental evidence and functional annotations.

### LiteFold/InterPro
- **Type**: Protein Annotation
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/InterPro

InterPro entries — integrated protein family, domain, repeat, and functional-site signatures across multiple member databases for sequence classification.

### LiteFold/FireProtDB
- **Type**: Protein Stability
- **Tags**: Biology, Medicine
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/FireProtDB

FireProtDB — curated thermostability mutation data for proteins, supporting protein-engineering and stability-prediction modelling.

### LiteFold/HumanProteinAtlas
- **Type**: Protein Annotation
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/HumanProteinAtlas

Human Protein Atlas — gene and protein expression across tissues, cells, organs, pathology, and subcellular locations for human proteins.

### LiteFold/Pfam
- **Type**: Protein Families
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/Pfam

Pfam — protein family and domain annotation built from curated multiple sequence alignments and profile HMMs, mirrored to the Hub as JSONL.

### LiteFold/IEDB
- **Type**: Immunology
- **Tags**: Biology, Medicine
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/IEDB

IEDB assay export — experimentally characterised B-cell, T-cell, and MHC-binding epitopes across infectious, allergic, autoimmune, and transplant settings.

### LiteFold/GO
- **Type**: Ontology
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/GO

Gene Ontology terms — structured controlled vocabulary for describing gene-product functions, biological processes, and cellular components across organisms.

### LiteFold/BFD
- **Type**: Protein Sequences
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/BFD

BFD source archive index — clustered UniProt + metagenomic protein sequences, the standard database for AlphaFold-style homology search and MSA construction.

### LiteFold/GOA
- **Type**: Ontology
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/GOA

Gene Ontology Annotation (UniProt) — high-quality evidence-coded GO annotations for UniProtKB proteins, RNAs, and protein complexes.

### LiteFold/protenix-data
- **Type**: Structure Prediction
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/protenix-data

Training data for Protenix — ByteDance’s open-source PyTorch reproduction of AlphaFold3 handling proteins, DNA, RNA, ligands, ions, and modifications.

### LiteFold/Evolutionary
- **Type**: MSA / Training Corpus
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/LiteFold/Evolutionary

Precomputed evolutionary sequence-alignment data (MMseqs2-style) packaged for the Hub — drop-in MSA features for protein-LM and structure-prediction pipelines.

### ratschlab/TCGA_virtual_spatial_transcriptomics_atlas
- **Type**: Spatial Transcriptomics
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/datasets/ratschlab/TCGA_virtual_spatial_transcriptomics_atlas

Population-scale virtual spatial transcriptomics atlas generated with DeepSpot-M by computationally augmenting TCGA retrospective histopathology archives — 28,664 whole-slide images from 10,865 patients across 32 cancer types.

Spatially resolved molecular maps covering 19k+ genes inferred from H&E images, enabling population-level studies of tissue biology for cancer research and AI-driven precision medicine.

### living-models/Botanic1-pretraining
- **Type**: Genomic Sequences
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/datasets/living-models/Botanic1-pretraining

The genomic sequence corpus used to pre-train the Botanic1 plant genome models — 7.2M training windows of 8,192 bp spanning 320 embryophyte species, roughly 59B effective base pairs.

Shards preserve original sequence case so the repeat soft mask survives, alongside source manifests, assembly catalogues, and an annotation quality audit for reproducible pre-training.

## Models (107)

### PhenoSeq
- **Type**: transcriptomics microscopy
- **Tags**: Biology
- **HuggingFace**: https://huggingface.co/Sentinal4D/PhenoSeq

PhenoSeq is a Gaussian diffusion model that generates scGPT RNA-seq embeddings conditioned on ViT-L microscopy imaging features. Given fluorescence microscopy images of a cell or well, it predicts a 512-dimensional scGPT embedding representing the transcriptomic state of individual cells — enabling image-to-transcriptome translation at single-cell resolution.

### Evo-2 40B
- **Type**: Genomic Foundation Model
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/arcinstitute/evo2_40b

40B-parameter DNA language model trained on 9.3 trillion nucleotides across all domains of life — zero-shot function prediction, variant effect scoring, and sequence generation.

### Evo-2 7B
- **Type**: Genomic Foundation Model
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/arcinstitute/evo2_7b

7B-parameter instruction-tuned DNA language model for gene function prediction, CRISPR guide design, and cross-species sequence analysis.

### STACK Large
- **Type**: Single-Cell Foundation Model
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/arcinstitute/Stack-Large

Large-scale single-cell transcriptomics foundation model supporting in-context learning across cell types and perturbation states.

### TEDDY
- **Type**: Single-Cell Biology
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/Merck/TEDDY

Transformer for Enabling Drug Discovery - foundation models trained on 116M single cells for genomics and drug discovery.

### NatureLM-audio
- **Type**: Audio-Language Model
- **Tags**: Biology, Ecology, Conservation, Earth Science
- **HuggingFace**: https://huggingface.co/EarthSpeciesProject/NatureLM-audio

First audio-language foundation model for bioacoustics - species classification, detection, and captioning of animal vocalizations.

### AVES2-BEATs
- **Type**: Bioacoustics Model
- **Tags**: Biology, Ecology, Conservation, Earth Science
- **HuggingFace**: https://huggingface.co/EarthSpeciesProject/esp-aves2-sl-beats-all

Self-supervised BEATs-based audio encoder trained on broad bioacoustic data for species detection, classification, and embedding across animal taxa.

### AQAffinity
- **Type**: Drug Discovery
- **Tags**: Chemistry, Medicine, Biology
- **HuggingFace**: https://huggingface.co/SandboxAQ/AQAffinity

Open-source protein-ligand binding affinity prediction model for drug discovery.

### MedGemma 1.5 4B
- **Type**: Medical AI
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/google/medgemma-1.5-4b-it

Multimodal medical AI model for medical imaging and clinical text understanding.

### MedGemma 27B
- **Type**: Medical AI
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/google/medgemma-27b-it

Large-scale instruction-tuned medical AI for radiology report generation, pathology image analysis, dermatology, and clinical question answering.

### AlphaGenome
- **Type**: Genomics
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/google/alphagenome-all-folds

Google DeepMind model predicting DNA regulatory features — gene expression, chromatin accessibility, and TF binding — at single-nucleotide resolution.

### MedSigLIP
- **Type**: Medical Imaging
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/google/medsiglip-448

Medical image-language model for visual understanding in healthcare.

### TxGemma 2B
- **Type**: Drug Discovery
- **Tags**: Medicine, Chemistry, Biology
- **HuggingFace**: https://huggingface.co/google/txgemma-2b-predict

Lightweight therapeutic prediction model for drug discovery tasks.

### TxGemma 9B Predict
- **Type**: Drug Discovery
- **Tags**: Medicine, Chemistry, Biology
- **HuggingFace**: https://huggingface.co/google/txgemma-9b-predict

Mid-size therapeutic prediction model for drug property prediction.

### TxGemma 9B Chat
- **Type**: Drug Discovery
- **Tags**: Medicine, Chemistry, Biology
- **HuggingFace**: https://huggingface.co/google/txgemma-9b-chat

Conversational therapeutic model for drug discovery with reasoning explanations.

### TxGemma 27B Predict
- **Type**: Drug Discovery
- **Tags**: Medicine, Chemistry, Biology
- **HuggingFace**: https://huggingface.co/google/txgemma-27b-predict

Large therapeutic prediction model achieving best-in-class performance on 66 tasks.

### TxGemma 27B Chat
- **Type**: Drug Discovery
- **Tags**: Medicine, Chemistry, Biology
- **HuggingFace**: https://huggingface.co/google/txgemma-27b-chat

Large conversational therapeutic model with advanced reasoning capabilities.

### Path Foundation
- **Type**: Pathology
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/google/path-foundation

Vision transformer for histopathology image embeddings - trained on 60M patches from TCGA.

### NTv3 650M
- **Type**: Genomics
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/InstaDeepAI/NTv3_650M_post

Multi-species genomics foundation model handling 1Mb context for functional track prediction.

### Nucleotide Transformer v2 500M
- **Type**: Genomics
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/InstaDeepAI/nucleotide-transformer-v2-500m-multi-species

500M multi-species DNA language model with improved tokenisation and benchmark performance across 18 genomic prediction tasks.

### Nucleotide Transformer 2.5B
- **Type**: Genomics
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/InstaDeepAI/nucleotide-transformer-2.5b-multi-species

2.5B-parameter DNA language model trained on 850 species genomes — state-of-the-art on promoter, enhancer, and splice site prediction tasks.

### ChatNT
- **Type**: Genomics
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/InstaDeepAI/ChatNT

8B multimodal conversational model for DNA, RNA, and protein tasks — instruction-following for sequence annotation, classification, and generation.

### Isoformer
- **Type**: Multi-Omics
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/InstaDeepAI/isoformer

Transformer model integrating DNA sequence, RNA expression, and protein context for isoform-level gene expression prediction.

### ESM2 650M
- **Type**: Protein Language Model
- **Tags**: Biology, Chemistry, Medicine
- **HuggingFace**: https://huggingface.co/facebook/esm2_t33_650M_UR50D

650M-parameter protein language model trained on UniRef50 — state-of-the-art embeddings for structure prediction, function annotation, and mutation effect scoring.

### Tahoe-x1
- **Type**: Single-Cell Biology
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/tahoebio/Tahoe-x1

Perturbation-trained single-cell foundation models (70M-3B) for cancer research and drug discovery.

### Tahoe-100M-SCVI
- **Type**: Single-Cell Biology
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/tahoebio/Tahoe-100M-SCVI-v1

scVI-based variational autoencoder trained on the full Tahoe-100M atlas of 100M+ single-cell profiles across 50 cancer lines and 1,100 drug perturbations.

### PeptiVerse
- **Type**: Peptide Design
- **Tags**: Biology, Chemistry, Medicine
- **HuggingFace**: https://huggingface.co/ChatterjeeLab/PeptiVerse

Foundation model for peptide design and analysis.

### CoLiPRI
- **Type**: Protein-Ligand Interaction
- **Tags**: Biology, Chemistry, Medicine
- **HuggingFace**: https://huggingface.co/microsoft/colipri

Contrastive learning model for protein-ligand interaction prediction.

### Phikon-v2
- **Type**: Pathology Foundation Model
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/owkin/phikon-v2

State-of-the-art histopathology vision foundation model trained with DINOv2 on 460K whole-slide images, achieving top performance on cancer subtyping and survival prediction.

### Phikon
- **Type**: Pathology Foundation Model
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/owkin/phikon

ViT-based pathology foundation model trained on TCGA and other large histopathology cohorts via self-supervised learning for cancer tissue representation.

### X-Cell
- **Type**: Single-Cell Perturbation Model
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/Xaira-Therapeutics/X-Cell

Diffusion-based model for predicting transcriptomic responses to CRISPRi perturbations at single-cell resolution, trained on the X-Atlas-Pisces dataset.

### p-IgGen
- **Type**: Antibody Language Model
- **Tags**: Biology, Medicine, Chemistry
- **HuggingFace**: https://huggingface.co/opig/p-IgGen

GPT-NeoX-based generative language model for antibody sequence design, trained on the Observed Antibody Space to generate diverse immunoglobulin heavy and light chains.

### OpenFold3
- **Type**: Protein Structure Prediction
- **Tags**: Biology, Medicine, Chemistry
- **HuggingFace**: https://huggingface.co/OpenFold/OpenFold3

Open replication of AlphaFold3 — predicts structures of proteins, nucleic acids, ligands, and their complexes for drug discovery and structural biology.

### MedSAM
- **Type**: Medical Image Segmentation
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/wanglab/medsam-vit-base

SAM ViT-Base finetuned on a large-scale dataset of CT, MRI, X-ray, ultrasound, and histology — a universal promptable foundation model for medical image segmentation.

### GO-GPT
- **Type**: Protein Function Model
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/wanglab/gogpt

Generative model that predicts Gene Ontology functional annotations directly from protein sequences — bringing LLM-style decoding to functional protein characterisation.

### OpenMed PharmaDetect
- **Type**: Biomedical NER
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/OpenMed/OpenMed-NER-PharmaDetect-SuperClinical-434M

Token-classification model for pharmaceutical entity recognition in clinical text — built on the SuperClinical 434M backbone for high-recall drug, dose, and regimen extraction.

### OpenMed BloodCancerDetect
- **Type**: Biomedical NER
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/OpenMed/OpenMed-NER-BloodCancerDetect-TinyMed-65M

Compact 65M token-classification model that identifies haematologic malignancy mentions (leukaemia, lymphoma, myeloma subtypes) in clinical and biomedical text.

### OpenMed ChemicalDetect
- **Type**: Biomedical NER
- **Tags**: Medicine, Chemistry, Biology
- **HuggingFace**: https://huggingface.co/OpenMed/OpenMed-NER-ChemicalDetect-ModernMed-149M

Chemical-entity NER over biomedical literature — identifies drug names, compounds, and chemical substances using the ModernMed 149M backbone.

### OpenMed SpeciesDetect
- **Type**: Biomedical NER
- **Tags**: Biology, Medicine
- **HuggingFace**: https://huggingface.co/OpenMed/OpenMed-NER-SpeciesDetect-ElectraMed-109M

Species-mention NER over biomedical literature — identifies organisms and taxonomic references using the ElectraMed 109M backbone.

### OpenMed DNADetect
- **Type**: Biomedical NER
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/OpenMed/OpenMed-NER-DNADetect-SuperMedical-125M

DNA-mention NER for biomedical text — extracts gene-level DNA sequence references and locus identifiers using the SuperMedical 125M backbone.

### OpenMed PathologyDetect
- **Type**: Biomedical NER
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/OpenMed/OpenMed-NER-PathologyDetect-TinyMed-135M

Pathology-finding NER over clinical and biomedical text — surfaces histopathological observations, lesion descriptions, and tissue-level abnormalities.

### OpenMed AnatomyDetect
- **Type**: Biomedical NER
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/OpenMed/OpenMed-NER-AnatomyDetect-ElectraMed-109M

Anatomical-entity NER for biomedical text — labels body parts, organ systems, and tissue references using the ElectraMed 109M backbone.

### OpenMed OncologyDetect
- **Type**: Biomedical NER
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/OpenMed/OpenMed-NER-OncologyDetect-MultiMed-568M

Oncology-focused NER that identifies cancer-type mentions, tumour grading, and staging language across clinical and biomedical literature.

### OpenMed OrganismDetect
- **Type**: Biomedical NER
- **Tags**: Biology, Medicine
- **HuggingFace**: https://huggingface.co/OpenMed/OpenMed-NER-OrganismDetect-TinyMed-82M

Organism-mention NER for biomedical text — broader than SpeciesDetect, also picking up genera, strains, and informal organism references.

### OpenMed DiseaseDetect
- **Type**: Biomedical NER
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/OpenMed/OpenMed-NER-DiseaseDetect-BioMed-335M

Disease-mention NER trained on the BioMed 335M backbone — recognises disease names, syndromes, and condition references in clinical and biomedical literature.

### OpenMed GenomicDetect
- **Type**: Biomedical NER
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/OpenMed/OpenMed-NER-GenomicDetect-PubMed-335M

Genomic-entity NER over PubMed-style text — labels genes, transcripts, and other genomic references for downstream knowledge extraction.

### OpenMed ProteinDetect
- **Type**: Biomedical NER
- **Tags**: Biology, Medicine
- **HuggingFace**: https://huggingface.co/OpenMed/OpenMed-NER-ProteinDetect-SuperClinical-141M

Protein-mention NER for biomedical and clinical text — extracts protein names, family references, and post-translational modification descriptors.

### OpenMed GenomeDetect
- **Type**: Biomedical NER
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/OpenMed/OpenMed-NER-GenomeDetect-ModernMed-149M

Genome-mention NER complementary to GenomicDetect — focuses on whole-genome and assembly-level references in biomedical text.

### BioReason-Pro SFT
- **Type**: Biological Reasoning Model
- **Tags**: Biology, Genomics, Scientific Reasoning
- **HuggingFace**: https://huggingface.co/wanglab/bioreason-pro-sft

Supervised fine-tuned variant of BioReason-Pro — multimodal biological reasoning over genomic variants and pathway data with chain-of-thought rationales.

### BioReason-Pro RL
- **Type**: Biological Reasoning Model
- **Tags**: Biology, Genomics, Scientific Reasoning
- **HuggingFace**: https://huggingface.co/wanglab/bioreason-pro-rl

RL-tuned variant of BioReason-Pro — reinforcement-learning fine-tuning over BioReason’s SFT base for sharper biological reasoning across KEGG pathways and variant data.

### NexaMass V3 Struct
- **Type**: Mass Spectrometry Model
- **Tags**: Chemistry, Biology, Scientific Reasoning
- **HuggingFace**: https://huggingface.co/AethronPhantom/NexaMass-V3-Struct

Self-supervised representation model for MS/MS spectra in metabolomics — learns molecular fingerprints to support compound identification and structure inference.

### MMPT-FM
- **Type**: Pharma Foundation Model
- **Tags**: Biology, Medicine, Chemistry
- **HuggingFace**: https://huggingface.co/Merck/MMPT-FM

Multi-modal pharma foundation model from Merck — integrates molecular and biological signals for drug discovery and target prediction.

### BioCLIP 2
- **Type**: Vision-Language Model
- **Tags**: Biology, Ecology, Conservation
- **HuggingFace**: https://huggingface.co/imageomics/bioclip-2

OpenCLIP-based foundation model for organismal biology — zero-shot species classification from photographs across the tree of life, trained on TreeOfLife-200M.

### BioEmu
- **Type**: Protein Dynamics Model
- **Tags**: Biology, Medicine, Chemistry
- **HuggingFace**: https://huggingface.co/microsoft/bioemu

Generative model for protein structural ensembles — emulates conformational dynamics for drug discovery and structural biology beyond static AlphaFold-style predictions.

### OneGenome-Rice
- **Type**: Genomic Foundation Model
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/ZhejiangLab/OneGenome-Rice

Mixtral-architecture genomic foundation model specialised for rice (Oryza sativa) — supports variant analysis, expression prediction, and breeding-relevant trait modelling.

### Genos 1.2B
- **Type**: Genomic Foundation Model
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/ZhejiangLab/Genos-1.2B

General-purpose 1.2B-parameter genomic foundation model spanning multiple organisms — base model for downstream gene-level and sequence-level prediction tasks.

### eva-rna
- **Type**: Transcriptomics Foundation Model
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/ScientaLab/eva-rna

Transformer foundation model producing sample-level and gene-level embeddings from RNA-seq profiles (bulk, microarray, pseudobulked single-cell) in human and mouse.

### Carbon-3B
- **Type**: Genomic Foundation Model
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/HuggingFaceBio/Carbon-3B

Flagship 3B-parameter genomic foundation model from Hugging Face Biology Research — Llama-architecture DNA language model trained for sequence generation, variant interpretation, and downstream genomic tasks.

### Carbon-500M
- **Type**: Genomic Foundation Model
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/HuggingFaceBio/Carbon-500M

Compact 500M-parameter Carbon variant — DNA language model lightweight enough for downstream fine-tuning on consumer hardware.

### Carbon-8B
- **Type**: Genomic Foundation Model
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/HuggingFaceBio/Carbon-8B

Largest 8B-parameter Carbon variant — top-of-line DNA language model for the most demanding genomic prediction, generation, and benchmark tasks.

### GENA-LM BERT large (T2T)
- **Type**: Genomic Language Model
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/AIRI-Institute/gena-lm-bert-large-t2t

BERT-large-style genomic foundation model trained on telomere-to-telomere human assemblies — supports variant interpretation, regulatory prediction, and downstream genomic tasks.

### GENA-LM BERT base (T2T)
- **Type**: Genomic Language Model
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/AIRI-Institute/gena-lm-bert-base-t2t

BERT-base-style genomic foundation model trained on T2T assemblies — lighter-weight backbone for genomic sequence understanding.

### ModernGENA large
- **Type**: Genomic Language Model
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/AIRI-Institute/moderngena-large

GENA-LM rebuilt on the ModernBERT architecture — larger, longer-context, RoPE-equipped genomic foundation model.

### ModernGENA base
- **Type**: Genomic Language Model
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/AIRI-Institute/moderngena-base

Compact ModernBERT-based GENA-LM variant — efficient genomic foundation model for downstream variant and expression tasks.

### HuatuoGPT-Vision 7B
- **Type**: Medical Vision-Language Model
- **Tags**: Medicine, Biology, Scientific Reasoning
- **HuggingFace**: https://huggingface.co/FreedomIntelligence/HuatuoGPT-Vision-7B

Medical multimodal LLM from the HuatuoGPT family — answers clinical questions over medical imagery (radiology, pathology, dermatology) using a 7B vision-language backbone.

### FlashPPI
- **Type**: Protein-Protein Interaction Model
- **Tags**: Biology, Medicine
- **HuggingFace**: https://huggingface.co/tattabio/flashppi

Fast protein-protein interaction prediction model — trained for high-throughput screening of interaction networks.

### gLM2 650M
- **Type**: Genomic Language Model
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/tattabio/gLM2_650M

650M-parameter genomic foundation model from Tatta Bio — trained on the OMG open-mixed-genomes corpus for sequence-level biological reasoning.

### OpenPhenom
- **Type**: Phenomics Foundation Model
- **Tags**: Biology, Medicine, Chemistry
- **HuggingFace**: https://huggingface.co/recursionpharma/OpenPhenom

Masked-autoencoder foundation model for high-content cell imaging — learns phenomic embeddings from millions of microscopy images for downstream drug-discovery and perturbation analysis.

### Stack-Large Aligned
- **Type**: Single-Cell Foundation Model
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/arcinstitute/Stack-Large-Aligned

Aligned variant of STACK-Large — single-cell foundation model fine-tuned for cross-batch consistency, supporting multi-study perturbation analysis and downstream alignment tasks.

### SE-600M
- **Type**: Single-Cell Foundation Model
- **Tags**: Biology, Genomics, Medicine
- **HuggingFace**: https://huggingface.co/arcinstitute/SE-600M

600M-parameter Single-cell Embeddings model from the STATE collection — generates embeddings for human single-cell RNA expression profiles to support cell-state and perturbation analysis.

### ESM3 (Small Open)
- **Type**: Protein Language Model
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/biohub/esm3-sm-open-v1

Open release of ESM3 — a multimodal generative protein language model trained jointly on sequence, structure, and function tokens for unified protein understanding and design.

### ESMC 6B
- **Type**: Protein Language Model
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/biohub/ESMC-6B

Flagship 6B-parameter ESMC protein language model — state-of-the-art protein embeddings for variant-effect prediction, protein engineering, and biology research, trained on billions of protein sequences.

### ESMC 600M
- **Type**: Protein Language Model
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/biohub/ESMC-600M

600M-parameter ESMC protein language model — provides digital representations of proteins for therapeutic engineering, variant-effect prediction, and transfer learning across protein tasks.

### ESMC 300M
- **Type**: Protein Language Model
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/biohub/ESMC-300M

300M-parameter ESMC protein language model — compact variant of the ESMC family for protein-embedding workflows, fits on a single consumer GPU.

### ESMC 600M (Dec 2024)
- **Type**: Protein Language Model
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/biohub/esmc-600m-2024-12

December 2024 release of the 600M-parameter ESMC protein language model — widely adopted checkpoint underpinning protein engineering and variant-effect workflows.

### ESMC 300M (Dec 2024)
- **Type**: Protein Language Model
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/biohub/esmc-300m-2024-12

December 2024 release of the 300M-parameter ESMC protein language model — compact, widely used checkpoint for protein-embedding pipelines.

### ESMFold2
- **Type**: Protein Structure Prediction
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/biohub/ESMFold2

State-of-the-art protein structure prediction and design model — defines a new frontier for speed and accuracy, with unprecedented success rates for protein binder and antibody generation.

### ESMFold2-Fast
- **Type**: Protein Structure Prediction
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/biohub/ESMFold2-Fast

Accelerated variant of ESMFold2 — optimised for high-throughput structure prediction across large protein sets while retaining strong accuracy.

### ESMFold2-Experimental
- **Type**: Protein Structure Prediction
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/biohub/ESMFold2-Experimental

Experimental ESMFold2 variant exploring new training recipes and architectural changes for protein structure prediction, design, and confidence estimation.

### ESMFold2-Experimental-Fast
- **Type**: Protein Structure Prediction
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/biohub/ESMFold2-Experimental-Fast

Fast variant of the experimental ESMFold2 line — combines accelerated inference with the research-recipe improvements of the experimental branch.

### Protenix-RNA
- **Type**: RNA Structure Prediction
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/LiteFold/protenix-rna

PyTorch RNA-structure prediction model in the Protenix lineage — biomolecular structure predictor trained on the LiteFold/PDB index, supporting RNA structure tasks.

### Promera
- **Type**: Biomolecular Structure Model
- **Tags**: Biology, Medicine
- **HuggingFace**: https://huggingface.co/bjing-mit/promera

Unified generative model for biomolecular structure prediction, filtering, and design (Jing, Bafna, Diaz, Klivans, Berger — MIT / UT Austin, bioRxiv 2026). Combines all-atom co-folding (surpasses OpenFold3-p2 and Boltz-2 on therapeutic categories) with confidence-based binder filtering and controllable design of miniproteins and nanobodies via masked-sequence prediction with optional epitope, paratope, and template constraints.

### Rhaister
- **Type**: Perturbation Prediction Model
- **Tags**: Biology, Medicine, Genomics
- **HuggingFace**: https://huggingface.co/tahoebio/Rhaister

Perturbation-prediction model from Tahoe Therapeutics — trained on the Emerald Bay drug screen plus Tahoe-100M, Replogle-Nadig CRISPR, and Parse PBMC cytokine differential-expression summaries to predict single-cell perturbation responses across drugs, genetic knockdowns, and cytokines.

### UNI2-h
- **Type**: Pathology Foundation Model
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/MahmoodLab/UNI2-h

Second-generation UNI vision transformer trained self-supervised on >100M histology patches — high-throughput pathology image-feature extractor for downstream classification, segmentation, and slide-level inference.

### UNI
- **Type**: Pathology Foundation Model
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/MahmoodLab/UNI

Self-supervised ViT trained on 100K+ H&E whole-slide images — original UNI checkpoint used as a general-purpose patch encoder for computational pathology workflows.

### CONCH
- **Type**: Pathology Vision-Language Model
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/MahmoodLab/CONCH

Contrastive vision-language model for pathology trained on 1.17M histology image–caption pairs — enables zero-shot classification, retrieval, and report-grounded analysis on H&E whole-slide images.

### CONCH v1.5
- **Type**: Pathology Vision-Language Model
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/MahmoodLab/conchv1_5

Updated CONCH checkpoint (v1.5) with strengthened image-side encoder and expanded pretraining corpus — drop-in upgrade for zero-shot and few-shot pathology tasks.

### TITAN
- **Type**: Slide-Level Pathology Model
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/MahmoodLab/TITAN

Slide-level pathology foundation model — self-supervised ViT producing whole-slide embeddings for cancer subtyping, biomarker prediction, and treatment-response modelling.

### KRONOS
- **Type**: Spatial Proteomics Model
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/MahmoodLab/KRONOS

Self-supervised ViT-S/16 for multiplexed-imaging / spatial proteomics — pretrained for feature extraction on highly multiplexed tissue images across imaging modalities and panels.

### SEAL
- **Type**: Pathology Foundation Model
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/MahmoodLab/SEAL

SEAL — pathology foundation model from the Mahmood Lab covering whole-slide image representation and downstream evaluation across cancer classification and biomarker tasks.

### Madeleine
- **Type**: Pathology Model
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/MahmoodLab/madeleine

Madeleine — Mahmood Lab pathology research release; companion model used in their open evaluation pipeline for histology analysis.

### HEST Tissue Segmentation
- **Type**: Tissue Segmentation
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/MahmoodLab/hest-tissue-seg

Tissue segmentation model accompanying the HEST-1k spatial-transcriptomics resource — separates tissue from background across H&E whole-slide images for downstream patch extraction.

### ABMIL (UNI2 features, pc108-24k)
- **Type**: Slide-Level MIL Classifier
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/MahmoodLab/abmil.base.uni_v2.pc108-24k

Attention-based multiple-instance learning classifier on UNI2 features — pan-cancer pc108-24k training; pairs with the UNI2-h patch encoder for slide-level inference.

### ABMIL (CONCH v1.5 features, pc108-24k)
- **Type**: Slide-Level MIL Classifier
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/MahmoodLab/abmil.base.conch_v15.pc108-24k

Attention-based MIL classifier on CONCH v1.5 features — pan-cancer pc108-24k training for slide-level subtyping and downstream pathology evaluation.

### ABMIL (UNI features, pc108-24k)
- **Type**: Slide-Level MIL Classifier
- **Tags**: Medicine, Biology
- **HuggingFace**: https://huggingface.co/MahmoodLab/abmil.base.uni.pc108-24k

Attention-based MIL classifier on UNI features — pan-cancer pc108-24k training; provides slide-level aggregation over the original UNI patch encoder.

### Nesso
- **Type**: Drug discovery
- **Tags**: Biology, Genomics, Chemistry
- **HuggingFace**: https://huggingface.co/recursionpharma/nesso

Open-source, coarse-grained co-folding model that significantly accelerates binding-affinity predictions.

### MarinDNA m5.1 1B base model
- **Type**: Genomics
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/marin-dna/marin-dna-exp135-m5.1

MarinDNA m5.1 is a 1.12B-parameter, nucleotide-level causal language model developed with Marin.

### LULA-1.1
- **Type**: drug discovery
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/omtx/lula-1.1

LULA-1.1 is a lightweight, sequence-only protein-ligand binding scorer from Om Therapeutics. It takes a protein amino-acid sequence and ligand SMILES and returns a binding score. There is no structure input, docking, or folding step.

### OpenDDE
- **Type**: Biomolecular foundation model
- **Tags**: Biology, Genomics, Chemistry
- **HuggingFace**: https://huggingface.co/aurekaresearch/OpenDDE

OpenDDE is an open-source, all-atom biomolecular foundation model that turns co-folding into a scalable engine for structure prediction, design, and optimization in drug discovery.

### DeepSpot-M
- **Type**: Spatial Transcriptomics Model
- **Tags**: Medicine, Biology, Genomics
- **HuggingFace**: https://huggingface.co/ratschlab/DeepSpotM

Multimodal deep-learning model that predicts spatially resolved gene expression directly from H&E whole-slide images — a LoRA adapter over the Midnight pathology foundation model, covering 19k+ genes. Used to generate the TCGA and HEST virtual spatial transcriptomics atlases for computational pathology and oncology research.

### Botanic0-S
- **Type**: Plant Genomic Foundation Model
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/living-models/Botanic0-S

Smallest of the first-generation Botanic plant genome models — a 4-layer encoder-only transformer pre-trained with masked language modelling on nuclear DNA from 43 phylogenetically diverse plant genomes, using a 6-mer DNA tokenizer.

### Botanic0-M
- **Type**: Plant Genomic Foundation Model
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/living-models/Botanic0-M

Mid-sized Botanic0 model with 10 transformer layers, trained on the same 43-genome plant corpus for regulatory element annotation, gene expression inference, and variant effect prediction.

### Botanic0-L
- **Type**: Plant Genomic Foundation Model
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/living-models/Botanic0-L

Largest first-generation Botanic model at 40 layers, approaching 1B parameters. Competitive with state-of-the-art genomic foundation models zero-shot and after fine-tuning across regulatory annotation, expression, and variant effect tasks.

### Botanic1-S
- **Type**: Plant Genomic Foundation Model
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/living-models/Botanic1-S

318M-parameter bidirectional Mamba-2 encoder for plant genomes, pre-trained on 314.6B tokens from 320 embryophyte species at single-nucleotide resolution with an 8,192-token context.

### Botanic1-M
- **Type**: Plant Genomic Foundation Model
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/living-models/Botanic1-M

688M-parameter BiMamba2 plant genome model with 52 blocks — bidirectional state-space scans without attention or positional embeddings, trained with masked language modelling and reverse-complement augmentation.

### Botanic1-L
- **Type**: Plant Genomic Foundation Model
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/living-models/Botanic1-L

Largest public Botanic1 model at 2.1B parameters across 72 BiMamba2 blocks, for embedding, zero-shot variant effect scoring, and fine-tuning on plant genomic prediction tasks.

### Botanic1-M SAE (k=64, codebook 8192)
- **Type**: Sparse Autoencoder
- **Tags**: Biology, Genomics
- **HuggingFace**: https://huggingface.co/living-models/Botanic1-M-sae-k64-codebook8192

BatchTopK sparse autoencoders over the residual stream of Botanic1-M, one per probed block, decomposing each base's 1024-dimensional hidden state into 8,192 learned features for interpretability work on plant genomic representations.

## Blog Posts (33)

### Eve Bio: Mapping the Pharmone Drug Interaction
- **Author**: hugging-science
- **Date**: 2025-01-01
- **Tags**: Medicine, Biology, Chemistry
- **Link**: https://huggingface.co/blog/hugging-science/eve-bio-mapping-the-pharmone-drug-interaction

Understanding drug interactions through AI-powered pharmacogenomics.

### PromoterGPT
- **Author**: hugging-science
- **Date**: 2025-01-01
- **Tags**: Biology, Genomics
- **Link**: https://huggingface.co/blog/hugging-science/promoter-gpt

AI-powered promoter sequence design and analysis.

### AI for Food Allergies
- **Author**: hugging-science
- **Date**: 2025-01-01
- **Tags**: Medicine, Biology
- **Link**: https://huggingface.co/blog/hugging-science/ai-for-food-allergies

Applying AI to understand and predict food allergies.

### GDP: Generative Design for Proteins
- **Author**: cgeorgiaw
- **Date**: 2025-01-01
- **Tags**: Biology, Chemistry
- **Link**: https://huggingface.co/blog/cgeorgiaw/gdp

Generative models for protein design and engineering.

### Making Antibody Embeddings and Predictions
- **Author**: ginkgo-datapoints
- **Date**: 2025-01-01
- **Tags**: Biology, Medicine, Biotechnology
- **Link**: https://huggingface.co/blog/ginkgo-datapoints/making-antibody-embeddings-and-predictions

How to create and use antibody embeddings for therapeutic applications.

### SAIR: Accelerating Pharma R&D with AI-Powered Structural Intelligence
- **Author**: SandboxAQ
- **Date**: 2025-09-06
- **Tags**: Chemistry, Medicine, Biology
- **Link**: https://huggingface.co/blog/SandboxAQ/sair-data-accelerating-drug-discovery-with-ai

How SandboxAQ's SAIR dataset of 1M+ protein–ligand structures is enabling AI-powered drug discovery with unprecedented structural coverage.

### ThermoGFN-IF for Catalysis
- **Author**: AmelieSchreiber
- **Date**: 2026-03-10
- **Tags**: Biology, Chemistry, Medicine
- **Link**: https://huggingface.co/blog/AmelieSchreiber/thermogfn-if

A protein sequence design model fine-tuned with GFlowNets for thermostable and kinetically-aware enzyme engineering.

### A New Era in Multistep Enzyme Design
- **Author**: AmelieSchreiber
- **Date**: 2024-10-16
- **Tags**: Biology, Chemistry
- **Link**: https://huggingface.co/blog/AmelieSchreiber/a-new-era-of-enzyme-engineering

Exploring generative AI approaches for designing multistep enzymatic pathways for biosynthesis and biocatalysis.

### A Guide to Designing New Functional Proteins
- **Author**: AmelieSchreiber
- **Date**: 2024-07-02
- **Tags**: Biology, Chemistry
- **Link**: https://huggingface.co/blog/AmelieSchreiber/protein-optimization-and-design

A comprehensive guide to improving protein function, stability, and diversity using generative AI and ESM-2.

### RFDiffusion Potentials
- **Author**: AmelieSchreiber
- **Date**: 2024-05-14
- **Tags**: Biology, Chemistry
- **Link**: https://huggingface.co/blog/AmelieSchreiber/rfdiffusion-potentials

Using RFDiffusion with custom guiding potentials to steer protein structure generation toward desired functional properties.

### Predicting the Effects of Mutations on Protein Function with ESM-2
- **Author**: AmelieSchreiber
- **Date**: 2023-12-13
- **Tags**: Biology, Genomics
- **Link**: https://huggingface.co/blog/AmelieSchreiber/mutation-scoring

Using ESM-2 protein language model embeddings to score and predict the functional impact of point mutations.

### Faster Persistent Homology Alignment and Protein Complex Clustering
- **Author**: AmelieSchreiber
- **Date**: 2023-11-30
- **Tags**: Biology, Mathematics
- **Link**: https://huggingface.co/blog/AmelieSchreiber/faster-pha

Accelerating persistent homology alignment with ESM-2 embeddings and persistence landscapes for protein complex clustering.

### Clustering Protein Complexes using Persistent Homology
- **Author**: AmelieSchreiber
- **Date**: 2023-11-29
- **Tags**: Biology, Chemistry
- **Link**: https://huggingface.co/blog/AmelieSchreiber/esm-ppi

Combining persistent homology with ESM-2 fine-tuning for protein–protein interaction network prediction and complex clustering.

### ESM-2 for Generating and Optimizing Peptide Binders
- **Author**: AmelieSchreiber
- **Date**: 2023-11-23
- **Tags**: Biology, Medicine
- **Link**: https://huggingface.co/blog/AmelieSchreiber/esm-interact

Generating and optimising peptide binders for target proteins using ESM-2 embeddings and directed evolution.

### Persistent Homology Alignment: Replacing Multiple Sequence Alignments
- **Author**: AmelieSchreiber
- **Date**: 2023-11-15
- **Tags**: Biology, Mathematics
- **Link**: https://huggingface.co/blog/AmelieSchreiber/plm-persistent-homology-msa-replacement

Replacing traditional multiple sequence alignments with ESM-2 embeddings and persistent homology for structure-aware protein comparison.

### In Silico Directed Evolution of Protein Sequences with ESM-2
- **Author**: AmelieSchreiber
- **Date**: 2023-11-13
- **Tags**: Biology, Chemistry
- **Link**: https://huggingface.co/blog/AmelieSchreiber/directed-evolution-with-esm2

Using ESM-2 and EvoProtGrad to simulate directed evolution in silico, optimising protein sequences for target properties.

### QLoRA for ESM-2 and Post Translational Modification Site Prediction
- **Author**: AmelieSchreiber
- **Date**: 2023-11-11
- **Tags**: Biology, Genomics
- **Link**: https://huggingface.co/blog/AmelieSchreiber/esm2-ptm

Applying QLoRA fine-tuning to ESM-2 for accurate prediction of post-translational modification sites across protein sequences.

### Estimating the Intrinsic Dimension of Protein Sequence Embeddings
- **Author**: AmelieSchreiber
- **Date**: 2023-10-18
- **Tags**: Biology, Mathematics
- **Link**: https://huggingface.co/blog/AmelieSchreiber/intrinsic-dimension-of-proteins

Measuring the intrinsic dimensionality of ESM-2 protein embeddings to understand the geometric structure of protein sequence space.

### Predicting Protein–Protein Interactions Using a Protein Language Model
- **Author**: AmelieSchreiber
- **Date**: 2023-10-15
- **Tags**: Biology, Chemistry
- **Link**: https://huggingface.co/blog/AmelieSchreiber/protein-binding-partners-with-esm2

Using ESM-2 embeddings and linear sum assignment to predict protein–protein binding partners at scale.

### ESMBind Ensemble Models
- **Author**: AmelieSchreiber
- **Date**: 2023-09-22
- **Tags**: Biology, Genomics
- **Link**: https://huggingface.co/blog/AmelieSchreiber/esmbind-ensemble

Ensemble methods for ESMBind models to improve binding site prediction accuracy and robustness across protein families.

### ESMBind: Low Rank Adaptation of ESM-2 for Protein Binding Site Prediction
- **Author**: AmelieSchreiber
- **Date**: 2023-09-15
- **Tags**: Biology, Genomics
- **Link**: https://huggingface.co/blog/AmelieSchreiber/esmbind

Fine-tuning ESM-2 with LoRA adapters to predict protein binding sites with high accuracy and parameter efficiency.

### A Comprehensive Introduction to AI for Proteins (2026)
- **Author**: tamarind.bio
- **Date**: 2026-01-01
- **Tags**: Biology, Chemistry, Medicine
- **Link**: https://www.tamarind.bio/blog/a-comprehensive-introduction-to-ai-for-proteins

A thorough primer on the state of AI for protein science — covering structure prediction, protein language models, generative design, and the full open-source model landscape.

### Boltz-2: State of the Art Structure and Binding Affinity Prediction
- **Author**: tamarind.bio
- **Date**: 2025-06-18
- **Tags**: Biology, Chemistry, Medicine
- **Link**: https://www.tamarind.bio/blog/boltz2-state-of-the-art-structure-and-binding-affinity-prediction

Boltz-2 outperforms AlphaFold3 on antibody-antigen interfaces and sets a new state of the art for protein-ligand binding affinity prediction.

### Boltzdesign1: Designing De Novo Binders to More Than Just Proteins
- **Author**: tamarind.bio
- **Date**: 2025-06-01
- **Tags**: Biology, Chemistry, Medicine
- **Link**: https://www.tamarind.bio/blog/boltzdesign1-small-molecule-rna-dna-protein-metal-binder-design

BoltzDesign1 extends de novo binder design beyond protein targets to small molecules, RNA, DNA, and metal ions.

### OpenFold3 and The Future of Protein Folding
- **Author**: tamarind.bio
- **Date**: 2025-04-01
- **Tags**: Biology, Chemistry
- **Link**: https://www.tamarind.bio/blog/openfold3-fully-open-alphafold3-alternative

OpenFold3 is a fully open-source, commercially available AlphaFold3 alternative backed by the OpenFold Consortium — enabling unrestricted biomolecular structure prediction.

### IntFold: A New Best Structure Prediction Protocol
- **Author**: tamarind.bio
- **Date**: 2025-03-01
- **Tags**: Biology, Chemistry
- **Link**: https://www.tamarind.bio/blog/intfold-a-new-state-of-the-art

IntFold establishes a new state-of-the-art protocol for biomolecular complex structure prediction, setting records across standard benchmarks.

### Chai-1r: AlphaFold3 Level Performance, Now Completely Open Source
- **Author**: tamarind.bio
- **Date**: 2025-02-01
- **Tags**: Biology, Chemistry, Medicine
- **Link**: https://www.tamarind.bio/blog/chai-1-alphafold3-level-performance-now-completely-open-source

Chai-1r achieves AlphaFold3-level accuracy on protein-protein and antibody-antigen complexes with fully open weights and no usage restrictions.

### Computational De Novo Design of Antibodies and Nanobodies
- **Author**: tamarind.bio
- **Date**: 2025-01-01
- **Tags**: Biology, Medicine, Chemistry, Biotechnology
- **Link**: https://www.tamarind.bio/blog/de-novo-antibody-nanobody-vhh-scfv-rfdiffusion

A practical guide to designing antibody VHHs and scFvs de novo using RFdiffusion and ProteinMPNN, from target epitope to validated sequence.

### Predicting Antibody Properties & Developability
- **Author**: tamarind.bio
- **Date**: 2025-01-01
- **Tags**: Biology, Medicine, Chemistry, Biotechnology
- **Link**: https://www.tamarind.bio/blog/predicting-antibody-properties-developability

ML approaches for predicting key biophysical properties of therapeutic antibody candidates — stability, solubility, and immunogenicity — before wet-lab validation.

### Are Mini Proteins the Next Antibodies?
- **Author**: tamarind.bio
- **Date**: 2025-01-01
- **Tags**: Biology, Medicine, Chemistry
- **Link**: https://www.tamarind.bio/blog/mini-protein-antibodies

Examining the therapeutic potential of computationally designed miniproteins as a next-generation alternative to traditional antibody drugs.

### Boltz-1: AlphaFold3 Level Performance, Truly Open Source
- **Author**: tamarind.bio
- **Date**: 2024-11-01
- **Tags**: Biology, Chemistry
- **Link**: https://www.tamarind.bio/blog/boltz-1-alphafold3-level-performance-truly-open-source-and-commercially-available

Boltz-1 from MIT achieves AlphaFold3-level accuracy on protein and protein-ligand structure prediction with no restrictions on commercial use or input types.

### Computational De Novo Miniproteins As Therapeutics
- **Author**: tamarind.bio
- **Date**: 2024-12-01
- **Tags**: Biology, Medicine, Chemistry
- **Link**: https://www.tamarind.bio/blog/computationaly-de-novo-minibinders-therapeutic-applications

How computationally designed de novo miniproteins and minibinders are being developed as a new class of targeted therapeutics.

### Computational Protein–Protein Interaction Screening
- **Author**: tamarind.bio
- **Date**: 2024-12-01
- **Tags**: Biology, Medicine, Chemistry
- **Link**: https://www.tamarind.bio/blog/ppi-screen

A practical guide to screening for protein–protein interactions (PPIs) as drug discovery targets using structure prediction and ML scoring.