rewirebio.iobenchmarks

Biological models

Understand model architectures, biological inputs, access requirements and the evidence from linked evaluations.

370 model records in release 2026-11-18-43dae0e812e5 (release manifest, gzip). Showing 121–144; page 6 of 16.

Evaluated configurations, methods and pipelines are listed separately. Names alone do not establish equivalent models or checkpoints.

  • FOAM

    Metabolomics

    Formula-constrained graph genetic algorithm that evolves candidate molecules, seeded from PubChem formula matches and DiffMS samples, and ranks them by ICEBERG-simulated spectral similarity to the observed MS/MS spectrum.

  • FrameDiff

    Protein structure

    Diffusion-based protein backbone generation method evaluated in Scaffold-Lab (released 2023-03).

  • FrameFlow

    Protein structure

    Flow-based protein backbone generation method evaluated in Scaffold-Lab (released 2023-10).

  • FramePool

    RNA and transcriptomes

    Neural network with frame pooling that predicts mean ribosome load, a proxy for translation efficiency, from 5' UTR sequences of any length.

  • FRIGID

    Metabolomics

    FRIGID diffusion-based molecular generation model from mass spectra.

  • GAT

    Protein function

    Graph attention network over the protein graph.

  • GCN

    Protein function

    Graph convolutional network over the protein graph.

  • GEARS

    Cells and tissues

    GEARS predicts transcriptional responses to single- and multi-gene perturbations from single-cell perturbation screens.

  • Gemini 3 Flash

    Cells and tissues

    Google reasoning model built on the Gemini 3 Pro foundation, with thinking levels that trade quality against cost and latency, taking text, image, audio and video input with up to one million tokens of context.

  • Gemini 3 Pro

    Cells and tissues

    Google's natively multimodal sparse mixture-of-experts reasoning model that takes text, images, audio and video with up to one million tokens of context and outputs text.

  • Gemini 3.1 Pro

    Cells and tissues

    Google's hosted large language model Gemini 3.1 Pro.

  • Gemini 3.8 Flash

    DNA and genomes

    Large language model served as gemini-3.8-flash over Vertex AI with an API key, which plans molecular cloning experiments when driven by an agent harness.

  • Gemini 4 Argon

    DNA and genomes

    Large language model served as gemini-4-argon over the Vertex AI global endpoint, which plans molecular cloning experiments when driven by an agent harness.

  • GENA-LM

    DNA and genomes

    Transformer DNA language models in BERT-base, BERT-large and BigBird variants.

  • GenBench CNN baseline

    DNA and genomes

    Convolutional baseline trained from scratch in GenBench.

  • Geneformer

    Cells and tissues

    Geneformer represents single-cell transcriptomes as ranked genes and learns contextual gene and cell representations.

  • GenePT

    Single-cell omics · Spatial omics

    Cell embeddings built from GPT text embeddings of gene descriptions, weighted by expression (GenePT-w variant).

  • GENERATOR

    DNA and genomes

    Generative transformer decoder DNA language model family with 6-mer tokens and long context (98 kb in the first release, pretrained on eukaryotic DNA; the second version adds prokaryote variants), used for sequence generation, zero-shot variant effect scoring and as a source of embeddings for fine-tuning.

  • GENERator

    DNA and genomes

    Generative transformer decoder DNA language models with 6-mer tokens, in eukaryote and prokaryote variants, that generate and embed long sequences (98 kb context in the first release, pretrained on 386 billion nucleotides of eukaryotic DNA).

  • GenerRNA

    RNA and transcriptomes

    GenerRNA autoregressive generative RNA language model.

  • GeNet

    Microbes and communities

    Convolutional (ResNet) taxonomic classifier with one-hot and positional embeddings of a fixed-length read.

  • Genie

    Protein structure

    Diffusion-based protein backbone generation method evaluated in Scaffold-Lab (released 2023-01).

  • Genie 3

    Protein structure

    Genie 3 generates protein designs through all-atom equivariant diffusion.

  • GenomeOcean

    DNA and genomes

    Generative transformer DNA model trained on co-assembled environmental metagenomes, whose embeddings and generated sequences capture phylogenetic and protein-coding structure.

This index reflects a dated catalogue, not an exhaustive census. Source checking does not mean independent reproduction; compare results only under compatible protocols, datasets and metrics.