Jun Li, Haozhe Guo, Chen Song
We propose a pipeline, MemConverter, for the conversion of soluble proteins to membrane proteins and vice versa, based on the prediction of membrane contact probability (MCP). By fine-tuning ProteinMPNN with a membrane protein data set to create MemProtMPNN, and integrating it with iterative structure refinement using AlphaFold2, our approach enables precise reprogramming of protein surface properties. The pipeline utilizes MCP-guided sequence fusion to direct the evolution of protein sequences toward the desired membrane or soluble environments. The pipeline demonstrates superior performance compared to existing methods, and molecular dynamics simulations validate that the designed proteins exhibit stable membrane integration or enhanced solubility, depending on the target localization. This pipeline introduces a new computational tool for engineering proteins to be localized in membranes or aqueous solutions.