Generating Multistate Conformations of P-type ATPases with a Conditional Diffusion Model

Area of Science:

• Computational Biology
• Structural Biology
• Artificial Intelligence

Background:

• Understanding membrane protein conformational states is vital for biological function.
• Accurately predicting these complex structural changes computationally remains challenging.

Purpose of the Study:

• To develop a novel computational approach for generating diverse and biologically relevant membrane protein conformations.
• To apply and validate this method on P-type ATPases, a key family of membrane transporters.

Main Methods:

• Utilized a conditional diffusion model integrating forward and backward diffusion processes.
• Incorporated state classifiers and conditioners to guide the generation of conformational states.
• Employed a graph neural network with membrane constraints, trained on experimental and simulation data.

Main Results:

• Successfully generated a wide spectrum of P-type ATPase conformations linked to distinct functional states.
• Demonstrated exceptional accuracy in capturing biologically relevant structural diversity.
• The model effectively controls the conformational state generation gradient.

Conclusions:

• The developed AI approach significantly advances the computational generation of membrane protein conformations.
• This method shows great promise for studying the dynamics of P-type ATPases and other membrane proteins.