PharmaVQA: A Retrieval-Augmented Visual Question Answering Framework for Molecular Representation via Pharmacophore Guided Prompts
Keywords: Pharmacophore, Visual Question Answering, Prompt Learning, Molecule Represenation Learning, Bilinear Attention Network, Multi-modal Retrieval
TL;DR: Pharmacophore Guided VQA Framework for Molecular Representation
Abstract: In drug discovery, molecular representation learning is vital for understanding and generating new drug-like molecules. The accurate representation of molecules facilitates drug candidate screening and the optimization of lead compounds. The vastness of chemical space challenges traditional drug design and relies on complex computations. The Pharmacophore is a functional group contained within a drug molecule, which binds to receptors or biological macromolecules to produce biological effects and reduce computations. Pharmacophore-guided representation of molecules, however, remains a significant challenge. To address this issue, we propose an improved deep learning-based model called PharmaVQA for retrieving pharmacophore-related information directly from molecule databases, allowing for a more targeted understanding of drug-like molecules. Through the use of Visual Question Answering (VQA) framework, PharmaVQA captures pharmacophore data, generates knowledge prompts, and enriches molecular representations. On 46 benchmark datasets, PharmaVQA has demonstrated superior performance in both molecular property prediction and drug-target interaction prediction. Additionally, the applicability of PharmaVQA in drug discovery has been validated on an FDA-approved molecule dataset, where the Top-20 predictions were analyzed in real-world studies, with the majority of them experimentally validated as potential ligands previously reported in the literature. Our assessment of PharmaVQA is that it is a powerful and useful tool for accelerating the development of AI-assisted drug discovery across a wide range of areas.
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 10230
Loading