Biological models

318 model records in release 2026-10-23-5347695500ed (release manifest, gzip). Showing 265–288; page 12 of 14.

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

• RoseTTAFold All-Atom

  Protein structure · Molecular interactions

  All-atom co-folding model evaluated on the LiPP benchmark set.

• RosettaFoldNA

  RNA and transcriptomes · Protein structure

  RoseTTAFoldNA (RF2NA): three-track network that predicts protein and nucleic acid structures and complexes.

• S2F

  Proteins and complexes

  Sequence-Structure Fitness model that feeds ESM-2-650M outputs as node features to a Geometric Vector Perceptron structure encoder to predict protein fitness zero-shot, scored in ProteinGym as a substitution baseline.

• S2F-MSA

  Proteins and complexes

  Ensemble of the Sequence-Structure Fitness model (S2F) with EVE alignment-based predictions by summing their z-scores, scored in ProteinGym as a zero-shot substitution baseline.

• S3F

  Proteins and complexes

  Sequence-Structure-Surface Fitness model that combines protein language model sequence representations with Geometric Vector Perceptron encoders of the backbone and surface to predict protein fitness zero-shot, scored in ProteinGym as a substitution baseline.

• S3F-MSA

  Proteins and complexes

  Ensemble of the Sequence-Structure-Surface Fitness model (S3F) with EVE alignment-based predictions by summing their z-scores, scored in ProteinGym as a zero-shot substitution baseline.

• SaProt

  Proteins and complexes

  Protein language model trained on about 40 million protein sequences and structures using a structure-aware vocabulary that combines residue tokens with Foldseek structure tokens, scored in ProteinGym as a zero-shot substitution baseline from sequence and structure.

• scBERT

  Cells and tissues · Single-cell omics

  Single-cell foundation model, published as a pretrained deep language model for cell type annotation (reference 4); not designed for perturbation prediction and used here with a linear decoder (main text paragraph 2).

• scCello

  Cells and tissues

  Single-cell foundation model pretrained with cell-ontology alignment losses.

• scELMo

  Single-cell omics · Spatial omics

  Cell embeddings built from large language model embeddings of gene descriptions, weighted by expression.

• scFoundation

  Cells and tissues

  scFoundation produces contextual cell and gene representations from gene-expression measurements.

• scGPT

  Cells and tissues

  scGPT learns representations of single-cell molecular measurements and supports task-specific adaptation.

• scGPT-spatial

  Single-cell omics · Spatial omics

  scGPT continually pretrained on spatial transcriptomics from the SpatialHuman30M corpus.

• scPRINT

  Biological networks · Single-cell omics

  Single-cell foundation model pretrained on CELLxGENE data for gene network inference and expression denoising.

• scVI

  Cells and tissues

  scVI models single-cell RNA counts with a probabilistic latent-variable model that accounts for observed covariates.

• SegmentNT

  DNA and genomes

  SegmentNT labels genomic elements at individual nucleotide positions using a pretrained DNA backbone.

• Sei

  DNA and genomes

  Supervised sequence model built from residual dilated convolutions and pretrained on 21,907 chromatin profiling datasets.

• SiteRM

  Proteins and complexes

  Independent-sites phylogenetic model that estimates an amino acid substitution rate matrix for each column of a multiple sequence alignment and uses it for variant effect prediction, scored in ProteinGym as a zero-shot substitution baseline.

• SPACE

  DNA and genomes

  Supervised DNA model that learns sequence representations by predicting genomic profiles such as chromatin accessibility across species, using species- and profile-specific mixture-of-experts layers.

• Spec2Mol

  Metabolomics

  Spec2Mol spectrum-to-molecule generation model.

• SpeciesLM

  DNA and genomes

  Species-aware DNA language model (SpeciesLM).

• SpliceAI

  DNA and genomes

  SpliceAI annotates sequence variants with predicted splice acceptor and donor changes.

• SpliceAI2

  DNA and genomes

  Deep learning model that predicts splice site usage, splice junction usage and complete transcripts from about 200 kb of genomic sequence, and scores variant effects on splicing.

• SpliceBERT

  DNA and genomes

  Pre-trained language model of precursor mRNA sequence.

  Record: gfm2-20261011-you2026-model-splicebert

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.