Synergistic integration of topology and machine learning for the predictions of protein-ligand binding affinities and mutation impacts

项目摘要 阻碍当前理解生物分子系统的根本挑战是它们的 巨大的复杂性、高维度和与其相关的超大数据集 几何建模和仿真。这些挑战要求我们采取创新的应对策略 海量的生物分子数据集。与几何不同，拓扑学提供了一种独特的工具 降维和数据简化。然而，传统的拓扑通常会产生 几何信息的过度缩减。持久同调是拓扑学的一个新分支 它能够连接传统的拓扑学和几何学，但却忽视了生物学 信息。基于Pi最近在生物分子拓扑数据分析方面的工作，这 该项目将探索如何将拓扑数据分析和机器学习集成到显著 改进目前对蛋白质-配体结合和突变影响的最新预测 建立在PI的初步研究中。这些改进将通过以下方式实现 开发嵌入物理的拓扑方法和高级深度学习 用于处理来自各种物理数据的异质生物分子数据集的体系结构 以及生物学方面的考量。最后，PI将建立强大的数据库和在线服务器 对于提议的预测。

Project Summary Fundamental challenges that hinder the current understanding of biomolecular systems are their tremendous complexity, high dimensionality and excessively large data sets associated with their geometric modeling and simulations. These challenges call for innovative strategies for handling massive biomolecular datasets. Topology, in contrast to geometry, provides a unique tool for dimensionality reduction and data simplification. However, traditional topology typically incurs with excessive reduction in geometric information. Persistent homology is a new branch of topology that is able to bridge traditional topology and geometry, but suffers from neglecting biological information. Built upon PI’s recent work in the topological data analysis of biomolecules, this project will explore how to integrate topological data analysis and machine learning to significantly improve the current state-of-the-art predictions of protein-ligand binding and mutation impact established in the PI’s preliminary studies. These improvements will be achieved through developing physics-embedded topological methodologies and advanced deep learning architectures for tackling heterogeneous biomolecular data sets arising from a variety of physical and biological considerations. Finally, the PI will establish robust databases and online servers for the proposed predictions.