Navigating chemical-linguistic sharing space with heterogeneous molecular encoding
8 affiliations
- 1Beijing Key Laboratory of Brain-inspired Spiking Large Models, School of Computer Science, Peking University, Beijing, China
- 2Guangdong Provincial Key Laboratory of AI for Science and Computing (AI4SC), School of Artificial Intelligence for Science, Shenzhen Graduate School, Peking University, Shenzhen, China
- 3School of Electronic and Computer Engineering, Shenzhen Graduate School, Peking University, Shenzhen, China
- 4Guangdong Provincial Key Laboratory of AI for Science and Computing (AI4SC), School of Artificial Intelligence for Science, Shenzhen Graduate School, Peking University, Shenzhen, China. [email protected]
- 5School of Electronic and Computer Engineering, Shenzhen Graduate School, Peking University, Shenzhen, China. [email protected]
- 6Beijing Key Laboratory of Brain-inspired Spiking Large Models, School of Computer Science, Peking University, Beijing, China. [email protected]
- 7Guangdong Provincial Key Laboratory of AI for Science and Computing (AI4SC), School of Artificial Intelligence for Science, Shenzhen Graduate School, Peking University, Shenzhen, China. [email protected]
- 8School of Electronic and Computer Engineering, Shenzhen Graduate School, Peking University, Shenzhen, China. [email protected]
Abstract
Chemical language models (CLMs) are prominent for their effectiveness in exploring chemical space and enabling molecular design and engineering. However, while exploring chemical-linguistic space, CLMs suffer from the semantic gap between natural language and molecular representations. This challenge is primarily due to the inherent modeling differences between molecules and texts: molecules operate unified modeling to learn chemical space, while natural language sequentially models their semantic space. Additionally, the limited availability of high-quality text-to-molecule datasets further exacerbates this challenge. To address the problem, we first verified the information bias in molecular representations from different perspectives. We then developed the Heterogeneous Molecular Encoding (HME) framework, a unified molecular encoder compressing the molecular features from fragment sequence, topology, and conformation with Q-learning. To better model chemical-linguistic space, we further constructed the MCMoD dataset, which contains over one million molecules with various conditions, including properties, fragments, and descriptions. Experimentally, HME promotes CLMs to achieve chemical-linguistic sharing space exploration from two aspects: (1) chemical space exploration with linguistic guidance, where HME achieves significant improvements (+37.8% FCD) for molecular design in multiple constraints, even in zero-shot learning scenarios; (2) linguistic space exploration with molecular guidance, where HME generates textual descriptions with high qualities (+11.6% BLEU) for molecules. These results highlight the precision of HME in handling multi-objective and cross-domain tasks, as well as its remarkable generalization capability on unseen task combinations. HME offers a new perspective on navigating the chemical-linguistic sharing space, advancing the potential of CLMs in both fundamental research and practical applications in chemistry.
Related
- Molecular engineering of metalloporphyrins and phthalocyanines for homogeneous and heterogeneous CO₂ electroreduction
- Encoding metal coordination pathways and nuclearity by molecular knotting
- Biofabricated Al₂O₃ nanoparticles as a recyclable heterogeneous catalyst for nicotinonitrile synthesis: antimicrobial activity, molecular docking and ADMET assessment.
- Galvanic Coupling Links Heterogeneous and Molecular Palladium Catalysis in Aerobic Alcohol Oxidation
- A Machine Learning Approach to logBB Prediction Combining Molecular Docking and Quantum-Chemical Descriptors
BibTeX
@article{lv2026navigating,
title = {{Navigating chemical-linguistic sharing space with heterogeneous molecular encoding}},
author = {Liuzhenghao Lv and Hao Li and Yu Wang and Zijun Chen and Zhiyuan Yan and Zongying Lin and Yuan Li and Li Yuan and Yonghong Tian},
journal = {Nature Communications},
year = {2026},
month = jul,
doi = {10.1038/s41467-026-74666-w},
url = {https://doi.org/10.1038/s41467-026-74666-w},
note = {CC BY-NC-ND},
}
RIS
TY - JOUR
TI - Navigating chemical-linguistic sharing space with heterogeneous molecular encoding
AU - Liuzhenghao Lv
AU - Hao Li
AU - Yu Wang
AU - Zijun Chen
AU - Zhiyuan Yan
AU - Zongying Lin
AU - Yuan Li
AU - Li Yuan
AU - Yonghong Tian
PY - 2026
DA - 2026/07/10
JO - Nature Communications
DO - 10.1038/s41467-026-74666-w
UR - https://doi.org/10.1038/s41467-026-74666-w
AB - Chemical language models (CLMs) are prominent for their effectiveness in exploring chemical space and enabling molecular design and engineering. However, while exploring chemical-linguistic space, CLMs suffer from the semantic gap between natural language and molecular representations. This challenge is primarily due to the inherent modeling differences between molecules and texts: molecules operate unified modeling to learn chemical space, while natural language sequentially models their semantic space. Additionally, the limited availability of high-quality text-to-molecule datasets further exacerbates this challenge. To address the problem, we first verified the information bias in molecular representations from different perspectives. We then developed the Heterogeneous Molecular Encoding (HME) framework, a unified molecular encoder compressing the molecular features from fragment sequence, topology, and conformation with Q-learning. To better model chemical-linguistic space, we further constructed the MCMoD dataset, which contains over one million molecules with various conditions, including properties, fragments, and descriptions. Experimentally, HME promotes CLMs to achieve chemical-linguistic sharing space exploration from two aspects: (1) chemical space exploration with linguistic guidance, where HME achieves significant improvements (+37.8% FCD) for molecular design in multiple constraints, even in zero-shot learning scenarios; (2) linguistic space exploration with molecular guidance, where HME generates textual descriptions with high qualities (+11.6% BLEU) for molecules. These results highlight the precision of HME in handling multi-objective and cross-domain tasks, as well as its remarkable generalization capability on unseen task combinations. HME offers a new perspective on navigating the chemical-linguistic sharing space, advancing the potential of CLMs in both fundamental research and practical applications in chemistry.
SN - 2041-1723
ER -