rewirebio.iobenchmarks

Biological models

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

374 model records in release 2026-11-21-c8bea25e5abb (release manifest, gzip). Showing 73–96; page 4 of 16.

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

  • DeepFoldRNA

    RNA and transcriptomes · Protein structure

    DeepFoldRNA: RNA structure prediction from deep self-attention networks with gradient-based folding.

  • DeepGene

    DNA and genomes

    Graph transformer DNA model that represents sequences with pan-genome and Minigraph graphs to capture population variation, with rotary position embeddings for sequences of varying length.

  • DeepGraphDTA

    Molecular interactions

    DeepGraphDTA: GraphDTA with a transposed-input 1D convolution protein encoder.

  • DeepMicrobes

    Microbes and communities

    Taxonomic classifier using 12-mer embeddings, bidirectional LSTM layers and self-attention.

  • DeepSequence

    Proteins and complexes

    Latent-variable deep generative model of a protein family's sequences, learned without labels from evolutionary sequence data, that predicts mutation effects; ProteinGym scores it as a zero-shot baseline from a multiple sequence alignment.

  • DeepSTARR

    DNA and genomes

    DeepSTARR enhancer activity CNN.

  • DeepVariant

    DNA and genomes

    Deep-learning variant caller that classifies pileup images with a convolutional network.

  • DGRNA

    RNA and transcriptomes

    DGRNA long-context RNA foundation model with bidirectional Mamba2 attention.

  • DiffDock-L

    Molecular interactions

    DiffDock-L places small-molecule ligands in protein structures using a diffusion docking model.

  • DiffMS

    Metabolomics

    DiffMS: formula-conditioned discrete diffusion model that generates molecules from spectra via a MIST encoder.

  • DNABERT

    DNA and genomes

    DNA language model with k-mer tokenisation.

    Record: gfm2-20261011-you2026-model-dnabert

  • DNABERT

    DNA and genomes

    DNABERT (first version) k-mer DNA language model.

    Record: lit-genomics-20261011-gfmbench-model-dnabert

  • DNABERT-2

    DNA and genomes

    DNABERT-2 learns DNA representations that can be adapted to genomic prediction tasks.

  • DNABERT-S

    DNA and genomes

    Species-aware DNA embedding model contrastively fine-tuned from DNABERT-2.

    Record: gfm2-20261011-buzkova2026-model-dnabert-s

  • DNABERT-S

    DNA and genomes

    DNABERT-2 encoder further trained with a contrastive mixup objective so that its sequence embeddings cluster DNA fragments by species without labels.

    Record: lit-genomics-20261011-geneb-model-dnabert-s

  • DNAGPT

    DNA and genomes

    GPT-style DNA models pretrained on human genomes (0.1B) or on genomes of many mammals (3B; over 200 billion base pairs), with extra sequence-order and GC-content tasks so they take both sequence and numerical inputs.

  • DNALONGBENCH CNN baseline

    DNA and genomes

    Convolutional neural network baseline trained per task by the DNALONGBENCH authors.

  • DPLM

    Protein function · Proteins and complexes

    DPLM is a discrete diffusion protein language model that continues training from ESM-2 and is used both to generate sequences and as an encoder.

  • DPLM-2

    Proteins and complexes · Protein structure

    DPLM-2 is a multimodal diffusion protein language model that generates sequence and structure, and designs a sequence for a supplied backbone.

  • DreaMS

    Metabolomics

    DreaMS learns molecular representations from tandem mass spectra using self-supervised learning.

  • DRfold

    RNA and transcriptomes · Protein structure

    DRfold: ab initio RNA structure prediction that combines end-to-end learning with deep geometrical potentials.

  • E1

    Protein function · Proteins and complexes

    E1 is a retrieval-augmented protein language model trained with homologous sequences in its context, which can also be run on a query sequence alone.

  • eccDNAMamba

    DNA and genomes

    Bidirectional Mamba state-space encoder for full-length extrachromosomal circular DNA up to 200 kb, trained with an augmentation that preserves circular topology.

  • Emb-Cos

    Metabolomics

    Contrastive model that learns a shared embedding space for molecules and MS/MS spectra with candidate InfoNCE to retrieve a metabolite's structure from its spectrum among candidate molecules, without pretrained encoders or score fusion.

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.