Multimodal fusion with relational learning for molecular property prediction

Abstract Graph-based molecular representation learning is essential for predicting molecular properties in drug discovery and materials science. Despite its importance, current approaches struggle with capturing the intricate molecular relationships and often rely on limited chemical knowledge durin...

Full description

Saved in:
Bibliographic Details
Main Authors: Zhengyang Zhou, Yunrui Li, Pengyu Hong, Hao Xu
Format: Article
Language:English
Published: Nature Portfolio 2025-07-01
Series:Communications Chemistry
Online Access:https://doi.org/10.1038/s42004-025-01586-z
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849238805534474240
author Zhengyang Zhou
Yunrui Li
Pengyu Hong
Hao Xu
author_facet Zhengyang Zhou
Yunrui Li
Pengyu Hong
Hao Xu
author_sort Zhengyang Zhou
collection DOAJ
description Abstract Graph-based molecular representation learning is essential for predicting molecular properties in drug discovery and materials science. Despite its importance, current approaches struggle with capturing the intricate molecular relationships and often rely on limited chemical knowledge during training. Multimodal fusion, which integrates information from graph and other data sources together, has emerged as a promising approach for enhancing molecular property prediction. However, existing studies explore only a narrow range of modalities, and the optimal integration stages for multimodal fusion remain largely unexplored. Furthermore, the reliance on auxiliary modalities poses challenges, as such data is often unavailable in downstream tasks. Here, we present MMFRL (Multimodal Fusion with Relational Learning), a framework designed to address these limitations by leveraging relational learning to enrich embedding initialization during multimodal pre-training. MMFRL enables downstream models to benefit from auxiliary modalities, even when these are absent during inference. We also systematically investigate modality fusion at early, intermediate, and late stages, elucidating their unique advantages and trade-offs. Using the MoleculeNet benchmarks, we demonstrate that MMFRL significantly outperforms existing methods with superior accuracy and robustness. Beyond predictive performance, MMFRL enhances explainability, offering valuable insights into chemical properties and highlighting its potential to transform real-world applications in drug discovery and materials science.
format Article
id doaj-art-e8b03b0ed20e45f6a9d6503e7b67bdba
institution Kabale University
issn 2399-3669
language English
publishDate 2025-07-01
publisher Nature Portfolio
record_format Article
series Communications Chemistry
spelling doaj-art-e8b03b0ed20e45f6a9d6503e7b67bdba2025-08-20T04:01:24ZengNature PortfolioCommunications Chemistry2399-36692025-07-018111010.1038/s42004-025-01586-zMultimodal fusion with relational learning for molecular property predictionZhengyang Zhou0Yunrui Li1Pengyu Hong2Hao Xu3Department of Computer Science, Brandeis UniversityDepartment of Computer Science, Brandeis UniversityDepartment of Computer Science, Brandeis UniversityDepartment of Medicine, Brigham and Women’s Hospital, Harvard Medical SchoolAbstract Graph-based molecular representation learning is essential for predicting molecular properties in drug discovery and materials science. Despite its importance, current approaches struggle with capturing the intricate molecular relationships and often rely on limited chemical knowledge during training. Multimodal fusion, which integrates information from graph and other data sources together, has emerged as a promising approach for enhancing molecular property prediction. However, existing studies explore only a narrow range of modalities, and the optimal integration stages for multimodal fusion remain largely unexplored. Furthermore, the reliance on auxiliary modalities poses challenges, as such data is often unavailable in downstream tasks. Here, we present MMFRL (Multimodal Fusion with Relational Learning), a framework designed to address these limitations by leveraging relational learning to enrich embedding initialization during multimodal pre-training. MMFRL enables downstream models to benefit from auxiliary modalities, even when these are absent during inference. We also systematically investigate modality fusion at early, intermediate, and late stages, elucidating their unique advantages and trade-offs. Using the MoleculeNet benchmarks, we demonstrate that MMFRL significantly outperforms existing methods with superior accuracy and robustness. Beyond predictive performance, MMFRL enhances explainability, offering valuable insights into chemical properties and highlighting its potential to transform real-world applications in drug discovery and materials science.https://doi.org/10.1038/s42004-025-01586-z
spellingShingle Zhengyang Zhou
Yunrui Li
Pengyu Hong
Hao Xu
Multimodal fusion with relational learning for molecular property prediction
Communications Chemistry
title Multimodal fusion with relational learning for molecular property prediction
title_full Multimodal fusion with relational learning for molecular property prediction
title_fullStr Multimodal fusion with relational learning for molecular property prediction
title_full_unstemmed Multimodal fusion with relational learning for molecular property prediction
title_short Multimodal fusion with relational learning for molecular property prediction
title_sort multimodal fusion with relational learning for molecular property prediction
url https://doi.org/10.1038/s42004-025-01586-z
work_keys_str_mv AT zhengyangzhou multimodalfusionwithrelationallearningformolecularpropertyprediction
AT yunruili multimodalfusionwithrelationallearningformolecularpropertyprediction
AT pengyuhong multimodalfusionwithrelationallearningformolecularpropertyprediction
AT haoxu multimodalfusionwithrelationallearningformolecularpropertyprediction