We propose to renew our R01 “Toward a deeper understanding of allostery and allotargeting by computational approaches” (GM139297) building upon the significant progress we made during the past funding term as well as breathtaking advances that took place in the applications of AI to structural biology in recent years. A major breakthrough has been the introduction of AlphaFold by Google DeepMind. The accessibility of > 200 million structures in the AlphaFold database, together with our elastic network model (ENM)-based that can efficiently characterize structural dynamics of proteins at the proteome scale, provided us with unique opportunities that we propose to pursue in the renewal. An overarching theme of our R01 is the development of tools that integrate physical-sciences-based models and methods with machine learning (ML) methods to advance biological and pharmacological sciences. Our Aim 1 tackles a fundamental problem that has been an obstacle to rational design of allosteric modulators, and which highlights the need for physical-sciences-based models that can better inform ML methods. In Aim 1, we seek to accurately model and distinguish the effects of positive and negative allosteric modulators (PAMs and NAMs) on the structure and function of their target proteins. The challenge has been to capture the subtle way local events at the small-molecule binding site elicit distinctive global responses in the proteins. Data on these subtle events are sparse, highlighting the need for physics-based models to tackle this challenge. We will develop a new methodology that combines two powerful tools, our ProDy interface that provides a robust description of global/cooperative mechanisms intrinsically accessible to allosteric proteins, and the weighted ensemble (WE) enhanced sampling method that successfully samples at full atomic resolution events not accessible by conventional MD, to generate pathways of subtle, local events that can be used to train ML models or, on