Adaptive Tensor Attention Networks with Cross-Domain Transfer for Drug-Target Interaction Prediction
Keywords: Domain Adaptive Prediction，Attention
TL;DR: Our framework introduces a novel DTI symplectic structure that captures the intrinsic geometry of drug-target interactions
Abstract: The prediction of drug-target interactions is fundamental to the advancement of drug discovery. We present a groundbreaking unified theory for Drug-Target Interaction prediction with Domain Adaptation (DTI-DA), seamlessly integrating concepts from quantum mechanics, differential geometry, and information theory. Our framework introduces a novel DTI symplectic structure that captures the intrinsic geometry of drug-target interactions, leading to a Quantum Optimal Transport theorem that provides a rigorous foundation for domain adaptation in the DTI context. We develop a quantum statistical mechanical formulation of DTI-DA, introducing DTI-preserving quantum channels and deriving a Quantum Wasserstein distance tailored to drug discovery applications. Our information-geometric perspective yields a Quantum Fisher-Rao metric for DTI, resulting in a quantum Cramer-Rao bound that establishes fundamental limits on DTI prediction accuracy. We propose a unified variational principle for DTI-DA, encompassing quantum and classical aspects, which leads to a novel algorithm based on geometric stochastic gradient Langevin dynamics. Furthermore, we extend classical statistical inference to the quantum domain, deriving a Quantum Rao-Blackwell theorem and a Quantum Bayesian Cramer-Rao bound specifically for DTI-DA. These theoretical advancements not only deepen our understanding of the DTI-DA problem but also suggest new algorithmic approaches with provable guarantees. Preliminary numerical experiments on quantum-inspired DTI-DA algorithms demonstrate significant improvements in prediction accuracy and domain adaptation capabilities compared to classical methods, particularly for challenging out-of-distribution scenarios in drug discovery.
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: 13677
Loading