GA4GH AI Registry

Models

Genomic and health AI models, each pointing at a full GA4GH Genomic AI Model Card.

Submit a new model

11 of 11 entries shown

Category
Keyword
Certification tier
GA4GH Genomic AI Safety Evaluation (GASE)
GA4GH Agent Runtime Risk Level

  • AlphaGenome

    DNA sequence model from Google DeepMind predicting regulatory and non-coding variant effects on gene expression, splicing, and chromatin state.

    UnsignedVariant effect prediction

  • ESM-variants (ESM-1v zero-shot variant scoring)

    Applies Meta FAIR's ESM-1v protein language model to zero-shot missense variant effect scoring, deployed as a public tool.

    UnsignedVariant effect prediction

  • ESM-2

    Protein language model family from Meta FAIR, learning structure and function representations from protein sequence.

    UnsignedProtein structure and function

  • Evo 2

    Genome modeling and design foundation model spanning DNA, RNA, and protein, from the Arc Institute.

    UnsignedGenomic foundation model

  • GatorTron

    Clinical-note language model family (345M-8.9B params) pretrained on de-identified University of Florida Health and MIMIC-III clinical text.

    UnsignedClinical text and note extraction

  • Geneformer

    Single-cell transcriptome foundation model for context-aware network-biology predictions, from the Broad Institute / Theodoris lab.

    UnsignedGenomic foundation model

  • HyenaDNA

    Long-range genomic foundation model, pretrained at single-nucleotide resolution on context lengths up to 1 million tokens.

    UnsignedGenomic foundation model

  • MedGemma

    Google's open-weight Gemma 3-based model family for medical text and multimodal clinical reasoning (dermatology, pathology, radiology, clinical text).

    UnsignedMultimodal genomic or health model

  • Nucleotide Transformer

    DNA foundation language model family (500M-2.5B params) pretrained on 850+ genomes, including the human 1000 Genomes collection.

    UnsignedGenomic foundation model

  • Prov-GigaPath

    Whole-slide digital pathology foundation model from Microsoft Research, trained on over one billion tiles from 170,000+ real-world cancer slides.

    UnsignedMedical and genomic imaging

  • scGPT

    Generative single-cell multi-omics foundation model pretrained on over 33 million cells, from the Wang lab (University Health Network / University of Toronto).

    UnsignedGenomic foundation model