Rosetta Machine Learning Models Accurately Classify Positional Effects of Thioamides on Proteolysis

• DOI: 10.1021/acs.jpcb.0c05981
• Publication Date: 2020-09-01T14:08:18Z
• Authors: Sam Giannakoulias; Sumant R. Shringari; Chunxiao Liu; Hoang Anh T. Phan; Taylor M. Barrett; John J. Ferrie; E. James Petersson
• Keywords: Thioamide; Proteolysis; Ramachandran plot

Thioamide substitutions of the peptide backbone have been shown to stabilize therapeutic and imaging peptides toward proteolysis. In order rationally design thioamide modifications, we developed a novel Rosetta custom score function classify positional effects on proteolysis in substrates serine cysteine proteases. Peptides interest were docked into proteases using FlexPepDock application Rosetta. Docked complexes modified contain thioamides parametrized through creation atom types based ab intio simulations. simulated, resultant structural provided features for machine learning classification as decomposed values function. An ensemble, majority voting model was be robust predictor previously unpublished holdout data. Theoretical control simulations with pseudo-atoms that modulate only one physical characteristic show differential prediction accuracy by optimized model. These pseudo-atom simulations, well statistical analyses full implicate steric binding being primarily responsible proteolytic resistance.