Enhancing PPB Affinity Prediction through Data Integration and Feature Alignment: Approaching Structural Model Performance with Sequences
Keywords: binding affinity, geometric deep learning, virtual screening
Abstract: One key step of protein drug development is the screening of protein-protein binding (PPB) affinity. The current mainstream screening method of PPB affinity is laboratory experiments, which are costly and time-consuming, making it difficult to quickly perform high-throughput screening. Various deep learning methods have been proposed to predict PPB affinity, but they are often limited by the availability of high-quality data and the compatibility of the algorithms with that data. In this work, we developed two AI models, PPBind-3D and PPBind-1D, to predict PPB affinity. PPBind-3D leverages structural information near the protein-protein binding interface to make its predictions. By employing monotonic neural network constrained multi-task learning, we effectively utilized heterogeneous affinity data from diverse wet lab experiments to expand the development dataset to over 23,000 samples, thereby enhancing the model's generalization capabilities. Additionally, PPBind-1D was developed using sequence data to address the lack of structural data in practical applications. During the training of PPBind-1D, we aligned it with PPBind-3D by incorporating an additional 42,108 no-affinity-label samples through an alignment approach. Finally, we demonstrated three application cases of our AI models in the virtual screening of protein drugs, illustrating that our models can significantly facilitate high-throughput screening.
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: 7438
Loading