Cheminformatics

BioMatrix: Towards a Comprehensive Biological Foundation Model Spanning the Modality Matrix of Sequences, Structures, and Language

Qizhi Pei, Zhimeng Zhou, Yi Duan, Yiyang Zhao, Wei Li, Han Guo, Liang He, Chengping Li, Chang-Yu Hsieh, Conghui He, Rui Yan, Lijun Wu

Featured July 3, 2026

Original Paper

This analysis was generated by SciGrove. Upload your own PDFs or enter a DOI — and get the same AI breakdown on any paper.

Get started

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

A new AI model called BioMatrix learns about tiny molecules and big proteins, their shapes, and what people say about them, all at once, like a super-smart translator for biology, helping discover new medicines faster.

In depth

The paper introduces BioMatrix, a novel multimodal foundation model that unifies diverse biological data types—molecular sequences and structures, protein sequences and structures, and natural language—into a shared discrete token space. This allows a single decoder-only architecture to natively understand and generate across all modalities and entity types using a unified next-token prediction objective, eliminating the need for external encoders or modality-specific output heads. The model achieves state-of-the-art performance on a wide range of biological tasks, particularly excelling in cross-modal and cross-entity reasoning.

Key Takeaways

1
BioMatrix is the first foundation model to simultaneously achieve native multimodality (sequences, structures, language) and broad entity coverage (molecules, proteins, interactions) within a single decoder-only architecture.
2
A unified tokenization scheme maps all biomolecular sequences, structures, and natural language into a shared discrete vocabulary, enabling uniform processing under a next-token prediction objective.
3
The model demonstrates state-of-the-art or competitive performance on 77 out of 80 diverse biological tasks, with significant gains on cross-modal and cross-entity challenges, validating the unified design.

Conceptual Flow

HIGH LEVEL

Methodology: Unifying Biological Data

The model turns all kinds of biology information, like molecule names, protein shapes, and written descriptions, into a single secret code that one smart computer brain can understand.

Molecule Name

Molecule Shape

Protein Name

Protein Shape

Science Text

Convert to Secret Code

Unified Code Stream

Results: Mastering Many Tasks

By understanding this single secret code, the model can then solve many different biology puzzles, like designing new molecules or predicting how proteins work, better than other specialized tools.

Unified Code Stream

Solve Many Puzzles

New Molecule Design

Protein Function

Drug Binding

Science Answers