Task-adaptive multimodal molecular representations for structure-sensitive property prediction
- 1Faculty of Applied Sciences, Macao Polytechnic University Macao 999078 China
- 2State Key Laboratory of Public Big Data, College of Computer Science and Technology, Guizhou University Guizhou 550025 China
- 3Department of Computer Science and Engineering, Chinese University of Hong Kong 999077 Hong Kong
Abstract
Structure-sensitive properties (SSPs), including activity cliffs and chirality-dependent properties, challenge molecular machine learning because small structural perturbations can cause abrupt property changes and invalidate smooth structure–property assumptions. Here, we present CAMF (Chirality- and Activity-cliff-aware Multimodal Framework), a task-adaptive framework that models SSPs through selective integration of complementary molecular evidence. To systematically evaluate this problem, we construct SSPBench, a benchmark spanning 77 conventional ADMET and physicochemical tasks together with activity-cliff and chirality-sensitive benchmarks. CAMF integrates molecular embeddings and expert-defined descriptors using random-forest-based feature selection and adaptive fusion, enabling property-specific prioritization of informative signals while reducing multimodal redundancy. Across ten baselines, CAMF achieves the best overall performance on SSP tasks, improving mean R2 by up to 29.5% on activity-cliff datasets and reducing MAE by up to 23.3% on 90 364 chiral molecules with TD-DFT-computed optical rotatory strengths. Ablation analyses show that these gains arise from task-adaptive multimodal integration rather than naive feature concatenation. More broadly, our results reveal that modality relevance is strongly task-dependent, with descriptors and 3D geometry becoming especially important in non-smooth property regimes. Case studies further support the interpretability and practical utility of CAMF in identifying activity-associated substructures and clinically relevant toxicity liabilities.
Related
- Hybrid-CatBoost: an empirical comparison of molecular representations for aqueous-solubility prediction on the TDC benchmark
- Structure-Independent Genetic Algorithm-Based Molecular Optimization Using Pocket-Aware Molecular Representations
- Transformer-based molecular fragment prediction using SMILES and DeepSMILES representations in a fragment-based drug discovery pipeline.
- DeMiR: Density-Matrix-Informed Representations for Molecular Property Prediction
- Scaffold-Controlled Evaluation of Molecular Representations for Antimalarial Activity Prediction
BibTeX
@article{lin2026taskadaptive,
title = {{Task-adaptive multimodal molecular representations for structure-sensitive property prediction}},
author = {Shaolong Lin and Silong Zhai and Shihang Wang and Xinke Zhan and Yuquan Li and Weihong Li and Li Qin and Lin Shi and Yanan Tian and Kai Xu and Kewei Zhou and Chunbin Gu and Huanxiang Liu and Xiaojun Yao},
journal = {Chemical Science},
year = {2026},
month = sep,
doi = {10.1039/d6sc02982e},
url = {https://doi.org/10.1039/d6sc02982e},
note = {CC BY 3.0},
}
RIS
TY - JOUR
TI - Task-adaptive multimodal molecular representations for structure-sensitive property prediction
AU - Shaolong Lin
AU - Silong Zhai
AU - Shihang Wang
AU - Xinke Zhan
AU - Yuquan Li
AU - Weihong Li
AU - Li Qin
AU - Lin Shi
AU - Yanan Tian
AU - Kai Xu
AU - Kewei Zhou
AU - Chunbin Gu
AU - Huanxiang Liu
AU - Xiaojun Yao
PY - 2026
DA - 2026/09/02
JO - Chemical Science
DO - 10.1039/d6sc02982e
UR - https://doi.org/10.1039/d6sc02982e
AB - Structure-sensitive properties (SSPs), including activity cliffs and chirality-dependent properties, challenge molecular machine learning because small structural perturbations can cause abrupt property changes and invalidate smooth structure–property assumptions. Here, we present CAMF (Chirality- and Activity-cliff-aware Multimodal Framework), a task-adaptive framework that models SSPs through selective integration of complementary molecular evidence. To systematically evaluate this problem, we construct SSPBench, a benchmark spanning 77 conventional ADMET and physicochemical tasks together with activity-cliff and chirality-sensitive benchmarks. CAMF integrates molecular embeddings and expert-defined descriptors using random-forest-based feature selection and adaptive fusion, enabling property-specific prioritization of informative signals while reducing multimodal redundancy. Across ten baselines, CAMF achieves the best overall performance on SSP tasks, improving mean R2 by up to 29.5% on activity-cliff datasets and reducing MAE by up to 23.3% on 90 364 chiral molecules with TD-DFT-computed optical rotatory strengths. Ablation analyses show that these gains arise from task-adaptive multimodal integration rather than naive feature concatenation. More broadly, our results reveal that modality relevance is strongly task-dependent, with descriptors and 3D geometry becoming especially important in non-smooth property regimes. Case studies further support the interpretability and practical utility of CAMF in identifying activity-associated substructures and clinically relevant toxicity liabilities.
SN - 2041-6520
SN - 2041-6539
ER -