Open infrastructure for biological AI
Find, evaluate, and build on the world's biological models.
Models, datasets, and benchmarks across proteins, genomics, single-cell, and drug discovery — every result reproducible, every artifact versioned and openly licensed.
128
Models
77
Datasets
45
Benchmarks
80
Papers
Recent models
Browse all models → single-cell/rbio
rbio1 is a 3B biology reasoning model from the Chan Zuckerberg Initiative, post-trained from Qwen2.5-3B-Instruct by reinforcement learning (GRPO) with virtual-cell models and the Gene Ontology as soft verifiers. It comes in about ten variants, one per verifier. The weights are distributed by CZI's Virtual Cells Platform, not on Hugging Face; this repository holds the code.
bio-nlp/gemma-3-27b-meditronfo-amharic
An Amharic-language variant of Gemma-3-27B-MeditronFO, from EPFL's LiGHT laboratory and the MeditronFO pipeline. Its model card is not published yet, so its licence and training data are not stated upstream.
bio-nlp/gemma-3-27b-meditronfo
A medical specialisation of Gemma-3-27B-it, fine-tuned on the Fully Open Meditron corpus (about 601k clinician-vetted medical QA and clinical-vignette examples) by EPFL's LiGHT laboratory. Part of MeditronFO, an auditable pipeline in which the corpus, the decontamination, the training and the evaluation are all published.
bio-nlp/eurollm-9b-meditronfo
A medical specialisation of EuroLLM-9B-Instruct, fine-tuned on the Fully Open Meditron corpus (about 601k clinician-vetted medical QA and clinical-vignette examples) by EPFL's LiGHT laboratory. Part of MeditronFO, an auditable pipeline in which the corpus, the decontamination, the training and the evaluation are all published.
bio-nlp/eurollm-22b-meditronfo
A medical specialisation of EuroLLM-22B-Instruct, fine-tuned on the Fully Open Meditron corpus (about 601k clinician-vetted medical QA and clinical-vignette examples) by EPFL's LiGHT laboratory. Part of MeditronFO, an auditable pipeline in which the corpus, the decontamination, the training and the evaluation are all published.
bio-nlp/olmo-2-32b-meditronfo
A medical specialisation of OLMo-2-32B, fine-tuned on the Fully Open Meditron corpus (about 601k clinician-vetted medical QA and clinical-vignette examples) by EPFL's LiGHT laboratory. Part of MeditronFO, an auditable pipeline in which the corpus, the decontamination, the training and the evaluation are all published.
Recent datasets
Browse all datasets →single-cell/10x-genomics-datasets
The public demonstration datasets 10x Genomics publishes for its platforms: single-cell gene expression, immune profiling, ATAC and multiome (Chromium), and spatial transcriptomics (Visium, Xenium), under CC BY 4.0. One is mirrored here in full: PBMC 3k, the filtered gene-barcode matrix of about 2,700 peripheral blood mononuclear cells used by the Seurat and Scanpy tutorials.
genomics/synapse-data-repository
Synapse is a collaborative, open-source research platform that allows teams to share data, track analyses, and collaborate.
bio-nlp/fully-open-meditron
The training corpus of Fully Open Meditron (EPFL, LiGHT laboratory): about 601k examples (~150M tokens) for supervised fine-tuning of clinical LLMs. Eight public medical QA datasets, plus three clinician-vetted synthetic parts: exam-style QA, QA grounded in 46,469 clinical practice guidelines, and open clinical vignettes. Decontaminated against the evaluation sets.
genomics/sars-cov-2-wuhan-hu-1
The 29,903 bp reference assembly of SARS-CoV-2 isolate Wuhan-Hu-1, NCBI RefSeq NC_045512.2, published by UCSC as the wuhCor1 assembly, together with the NCBI gene annotation on the same coordinates.
Reproducible
Every benchmark score traces to a pinned dataset, config, and commit.
Versioned
Immutable artifact revisions with full lineage between models and data.
Openly licensed
Clear, machine-readable licenses on every model and dataset.