Lipid Contact Probability - an Essential and Predictive Property for the Structural and Functional Studies of (Membrane) Proteins

A significant part of the membrane protein surface is stably exposed to hydrophobiccore lipid bilayers rather than aqueous environment. Surprisingly, this importinformation rarely considered for structure predictions. Here we propose anew property, namely ‘Lipid Contact Probability’ or LCP short, explicitlydescribe likelihood an amino acid a given sequence in direct contact with thehydrophobic core bilayers. We new method predict fromsequence by using machine learning-based model DCRNN and molecular dynamicssimulation results. With our prediction, demonstrate that mapprediction accuracy can be significantly improved ResNet-based model, forboth proteins soluble proteins. attribute complimentaryusage solvent accessibility describing exposing properties theprotein acids. also that, prediction usedto identify not only membrane-embedding proteins, but membrane-anchoringproteins, as verified dynamics simulation, providing efficient way forthe localization function from sequences.