(469i) Scientific Benchmarks Guide Energy Function Improvements for Membrane Protein Modeling and Design

Membrane proteins partner with the surrounding lipid environment to perform key functions including signaling, transport, and catalysis. They constitute 30% of all proteins and are targets for over 60% of pharmaceuticals on the market. However, our ability to predict and design membrane protein structures has been delayed by limited computational resources. In this work, we engineered an enhanced implicit membrane based on IMM1 (Lazaridis, 2003) for representing the heterogeneous lipid bilayer. Our model advances three aspects of IMM1: (1) we developed parameters to characterize different lipid compositions, (2) we derived new water-to-bilayer transfer free energies from thermodynamics experiments, and (3) we expanded IMM-pore to adapt to oblong aqueous pores and cavities. This updated solvation model was then integrated into the latest Rosetta all-atom energy function. To validate and optimize the model, we designed and iteratively performed fifteen scientific benchmark tests that leverage thermodynamic measurements from the literature and medium resolution X-Ray structures. These testing cycles were used to optimize reference energies, weights, and overall prediction and design performance. Our work is the first major update to the Rosetta all-atom membrane energy function since 2007. We meet the challenge of sparse validation data for membrane proteins through the design of diverse, fast-running benchmarks. The updated energy function will advance the accuracy of membrane protein modeling and design, while also establishing a framework for future membrane energy function developments.