Lipidomic Profile Reconstruction of Therapeutic Membrane Targets Using Physics‐Based Optimization with Limited Activity Data

Membrane-targeting peptide motifs recognize characteristic features of biological membrane targets through specific interactions with their lipid composition. To elucidate the complex relationship between binding sequences and lipid composition, an alternative computational framework to leverage limited sequence activity data by combining genetic algorithms with coarse-grained molecular dynamics simulations is proposed. As demonstrated through evolutionary optimization of model membranes containing mammalian lipid compositions, four true-positive and four false-positive antimicrobial peptide sequences contain sufficient signal to tune membrane properties that maximize the separation in insertion behavior between true positives (deep insertion) and false positives (shallow insertion) in coarse-grained molecular dynamics simulations. This yields a robust discriminative performance comparable to that of large data-driven classification methods on the balanced, small dataset tested. This finding reveals how subtle differences in lipid composition precisely control the selective binding of membrane-associated proteins, thereby regulating their trafficking, aggregation, and function within living cells. The used methodology reconstructs lipidomic profiles from limited sequence data, uniquely revealing targeting patterns for therapeutic peptides and the selectivity mechanisms of membrane-associated proteins.