Adaptive Drug Interaction Prediction via Enhanced Graph Representation Learning
Keywords: Domain-Aligned，Transfer Learning，Drug-Target Interaction
TL;DR: We introduce GraphPharmNet, a novel architecture that operates on DDI-DA bundles.
Abstract: This paper presents a groundbreaking theoretical framework for drug-drug interaction (DDI) prediction that seamlessly integrates domain adaptation (DA) techniques with advanced mathematical concepts. We introduce GraphPharmNet, a novel architecture that operates on DDI-DA bundles, leveraging gauge-equivariant geometric deep learning to capture the intricate structure of drug interactions across domains. Our approach reformulates the DDI prediction problem using the language of differential geometry, optimal transport, and symplectic geometry, viewing domain adaptation as a Hamiltonian flow on a statistical manifold. We develop a cohomological interpretation of domain invariance, characterizing robust DDI prediction features through the lens of persistent homology and sheaf theory. The domain adaptation process is analyzed using a geometric renormalization group framework, revealing a profound connection between the DDI-DA bundle's geometry and the emergence of domain-invariant predictive features. We further elucidate the spectral properties of the DDI-DA Laplacian, providing insights into the topological stability of domain adaptation in DDI prediction. Extensive experiments on benchmark datasets demonstrate that GraphPharmNet significantly outperforms existing methods, particularly in scenarios with limited data or when transferring knowledge across disparate domains. Our results highlight the power of this unified mathematical framework in capturing complex drug interactions and adapting to new domains, paving the way for more accurate, robust, and interpretable DDI prediction models. This work not only advances the field of computational drug discovery but also establishes a rigorous theoretical foundation for domain adaptation in graph-structured data, with potential applications across a wide range of scientific disciplines. Our anonymous github link: \textbf{https://anonymous.4open.science/r/GraphPharmNet-C9D9}
Primary Area: transfer learning, meta learning, and lifelong learning
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.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
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: 13687
Loading