Generating Multistate Conformations\nof P‑type ATPases with a\nConditional Diffusion Model

Understanding\nand predicting the diverse conformational states\nof membrane proteins is essential for elucidating their biological\nfunctions. Despite advancements in computational methods, accurately\ncapturing these complex structural changes remains a significant challenge.\nHere, we introduce a computational approach to generate diverse and\nbiologically relevant conformations of membrane proteins using a conditional\ndiffusion model. Our approach integrates forward and backward diffusion\nprocesses, incorporating state classifiers and additional conditioners\nto control the generation gradient of conformational states. We specifically\ntargeted the P-type ATPases, a critical family of membrane transporters,\nand constructed a comprehensive data set through a combination of\nexperimental structures and molecular dynamics simulations. Our model,\nincorporating a graph neural network with specialized membrane constraints,\ndemonstrates exceptional accuracy in generating a wide range of P-type\nATPase conformations associated with different functional states.\nThis approach represents a meaningful step forward in the computational\ngeneration of membrane protein conformations using AI and holds promise\nfor studying the dynamics of other membrane proteins.