AtlasFold: Protein structure prediction with metagenomic-scale language models
- 1Korea Advanced Institute of Science and Technology (KAIST)
Abstract
Protein language models (PLMs) trained on evolutionary sequences learn representations that encode protein structure, enabling direct structure prediction without multiple-sequence alignments (MSAs). Here we present the Atlas model family, an open and trainable system spanning protein language modeling, monomer folding, and protein-complex prediction. AtlasLM-3B is a 3B-scale language model trained with masked language modeling on approximately 1.56 billion sequences, including metagenomic data, and outperforms the similarly sized ESM2-3B in unsupervised contact prediction. Building on these representations, AtlasFold predicts all-atom protein structures and achieves state-of-the-art accuracy among PLM-based folding models. Fine-tuning AtlasFold for protein-complex prediction produces AtlasFold-Multimer (AtlasFold-M), whose antibody-antigen prediction performance is comparable to that of AlphaFold3 and ESMFold2. This protein-specific folding architecture enables fast, memory-efficient inference with AtlasFold and AtlasFold-M. By releasing the training code and data, stage checkpoints, and model weights under the MIT License, we provide a foundation for advancing PLM-based protein structure prediction.
Related
- Sequence-only prediction of VHH nanobody epitopes and paratopes on HER2 with a frozen protein language model ( ESM2) and a structure-derived residue prior
- Limitations of general protein language models and public TCRpMHC data in training specificity prediction models
- Protein Language Model-Conditioned Graph Neural Networks for Multitask GPCR Ligand Activity Prediction
- SurfGraphPro: Integrating Protein Language Models with Geometric Deep Learning On Coarse Protein Surfaces for Binding Site Prediction
- AltereGO: Prediction of Protein Function by Multimodal Fusion of Protein Language Models.
BibTeX
@misc{seo2026atlasfold,
title = {{AtlasFold: Protein structure prediction with metagenomic-scale language models}},
author = {S. Seo and H. Kim and S. Moon and W. Y. Kim and Team KAIST},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.04.749352},
url = {https://www.biorxiv.org/content/10.64898/2026.09.04.749352v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - AtlasFold: Protein structure prediction with metagenomic-scale language models
AU - S. Seo
AU - H. Kim
AU - S. Moon
AU - W. Y. Kim
AU - Team KAIST
PY - 2026
DA - 2026/09/04
PB - bioRxiv
DO - 10.64898/2026.09.04.749352
UR - https://www.biorxiv.org/content/10.64898/2026.09.04.749352v1
AB - Protein language models (PLMs) trained on evolutionary sequences learn representations that encode protein structure, enabling direct structure prediction without multiple-sequence alignments (MSAs). Here we present the Atlas model family, an open and trainable system spanning protein language modeling, monomer folding, and protein-complex prediction. AtlasLM-3B is a 3B-scale language model trained with masked language modeling on approximately 1.56 billion sequences, including metagenomic data, and outperforms the similarly sized ESM2-3B in unsupervised contact prediction. Building on these representations, AtlasFold predicts all-atom protein structures and achieves state-of-the-art accuracy among PLM-based folding models. Fine-tuning AtlasFold for protein-complex prediction produces AtlasFold-Multimer (AtlasFold-M), whose antibody-antigen prediction performance is comparable to that of AlphaFold3 and ESMFold2. This protein-specific folding architecture enables fast, memory-efficient inference with AtlasFold and AtlasFold-M. By releasing the training code and data, stage checkpoints, and model weights under the MIT License, we provide a foundation for advancing PLM-based protein structure prediction.
ER -