All-Atom Structure Tokenizer: AtomLM-9b for Biomolecular Structure Prediction
- 1Scinetics Company
Abstract
We develop an All-atom Structure Tokenizer that compresses biomolecular geometry into discrete structural tokens and achieves high-accuracy all-atom reconstruction. A hierarchical geometric encoder and finite scalar quantization map chemical units to a shared vocabulary of 8,192 tokens, while a joint coordinate de-tokenizer reconstructs backbone, side-chain, nucleic-acid, and ligand heavy atoms. The tokenizer achieves all-atom lDDT scores of 0.832 and 0.835 on CASP14 and CASP15, respectively. Building on this tokenizer, we train AtomLM-9b, an autoregressive language model based on Qwen3.5-9B that predicts structural tokens from molecular identities and connectivity and decodes them into complete atomic structures. We demonstrate, for the first time, protein structure prediction performance comparable to AlphaFold 3 with an all-atom-tokenizer-based structural language model: on the FoldBench protein-protein task, AtomLM-9b achieves a success rate of 75.01%, compared with the published AlphaFold 3 result of 72.93%. These results establish discrete all-atom tokenization as an effective interface between high-fidelity atomic reconstruction and language-model-based structure prediction.
Related
- RGI-Toolkit: Differentiable Restraints for Controllable Biomolecular Structure Prediction
- Boltz2-Notebook: An Interactive Google Colab Platform for Diffusion-Based Biomolecular Structure and Binding Affinity Prediction using the Boltz2 model.
- NuFold RNA Structure Database: Large-Scale RNA Tertiary Structure Prediction with Confidence Metrics
- Repurposing PeTriBERT for Protein Protein Interaction Prediction with Sequence Structure Fusion
- Agent-driven Model Development for RNA 3D Structure Prediction
BibTeX
@misc{yan2026allatom,
title = {{All-Atom Structure Tokenizer: AtomLM-9b for Biomolecular Structure Prediction}},
author = {J. Yan and D. Xie and Z. Wang and M. Xu and Z. Zhang},
howpublished = {bioRxiv},
year = {2026},
month = nov,
doi = {10.64898/2026.11.03.763913},
url = {https://www.biorxiv.org/content/10.64898/2026.11.03.763913v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - All-Atom Structure Tokenizer: AtomLM-9b for Biomolecular Structure Prediction
AU - J. Yan
AU - D. Xie
AU - Z. Wang
AU - M. Xu
AU - Z. Zhang
PY - 2026
DA - 2026/11/03
PB - bioRxiv
DO - 10.64898/2026.11.03.763913
UR - https://www.biorxiv.org/content/10.64898/2026.11.03.763913v1
AB - We develop an All-atom Structure Tokenizer that compresses biomolecular geometry into discrete structural tokens and achieves high-accuracy all-atom reconstruction. A hierarchical geometric encoder and finite scalar quantization map chemical units to a shared vocabulary of 8,192 tokens, while a joint coordinate de-tokenizer reconstructs backbone, side-chain, nucleic-acid, and ligand heavy atoms. The tokenizer achieves all-atom lDDT scores of 0.832 and 0.835 on CASP14 and CASP15, respectively. Building on this tokenizer, we train AtomLM-9b, an autoregressive language model based on Qwen3.5-9B that predicts structural tokens from molecular identities and connectivity and decodes them into complete atomic structures. We demonstrate, for the first time, protein structure prediction performance comparable to AlphaFold 3 with an all-atom-tokenizer-based structural language model: on the FoldBench protein-protein task, AtomLM-9b achieves a success rate of 75.01%, compared with the published AlphaFold 3 result of 72.93%. These results establish discrete all-atom tokenization as an effective interface between high-fidelity atomic reconstruction and language-model-based structure prediction.
ER -