Thermodynamics-inspired Structure Hallucination for Protein-protein Interaction Modeling

23 Sept 2023 (modified: 11 Feb 2024)Submitted to ICLR 2024EveryoneRevisionsBibTeX
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.
Keywords: Protein-protein Interaction, Mutation Effect Prediction
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: Modeling protein-protein interactions (PPI) represents a central challenge within the field of biology, and accurately predicting the consequences of mutations in this context is crucial for various applications, such as drug design and protein engineering. Recent advances in deep learning (DL) have shown promise in forecasting the effects of such mutations. However, the effectiveness of these models is hindered by two primary constraints. First and foremost, obtaining the structures of mutant proteins is a persistent challenge, as they are often elusive to acquire. Secondly, interactions take place dynamically, but thermodynamics is rarely integrated into the DL architecture design. To address these obstacles, we present a novel framework known as Refine-PPI, which incorporates two key enhancements. On the one hand, we introduce a structure refinement module that is trained by a mask mutation modeling (MMM) task on available wide-type structures and then is transferred to hallucinate the inaccessible mutant protein structures. Additionally, we employ a new kind of geometric networks to capture the dynamic 3D variations and encode the uncertainty associated with PPI. Through comprehensive experiments conducted on the established benchmark dataset SKEMPI, our results substantiate the superiority of the Refine-PPI framework. These findings underscore the effectiveness of our hallucination strategy to address the absence of mutant protein structure and hope to shed light on the prediction of the free energy change.
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: 7154
Loading