Integrating\nProtein Language Model and Molecular Dynamics\nSimulations to Discover Antibiofouling Peptides

Antibiofouling peptide materials prevent the nonspecific\nadsorption\nof proteins on devices, enabling them to perform their designed functions\nas desired in complex biological environments. Due to their importance,\nresearch on antibiofouling peptide materials has been one of the central\nsubjects of interfacial engineering. However, only a few antibiofouling\npeptide sequences have been developed. This narrow scope of antibiofouling\npeptide materials limits their capacity to adapt to the broad spectrum\nof application scenarios. To address this issue, we searched for antibiofouling\npeptides in the vast sequence pool of the microbiome library using\na combination of deep learning-based high-throughput search and molecular\ndynamics (MD) simulations. A random forest-based model with an ensemble\nof ten independent classifiers was developed. Each classifier was\ntrained by prompt-tuning the foundational protein language model Evolution\nScaling Modeling version 2 (ESM2) on a distinct training data set.\nWe constructed the databases containing the same amount of antibiofouling\nand biofouling peptide sequences to attenuate the bias of the existing\ndatabases. MD simulations were conducted to investigate the interfacial\nproperties of six selected peptide candidates and their interactions\nwith a lysozyme protein. Two known antibiofouling peptides, (glutamic\nacid (E)-lysine (K))<sub>15</sub> and (EK-proline (P))<sub>10</sub>, and one known fouling peptide, (glycine)<sub>30</sub>, were used\nas the reference. The MD simulation results indicate that five of\nthe six peptides present the potential to resist biofouling. Our research\nimplies that deep learning and molecular simulations can be integrated\nto discover functional peptide materials for interfacial applications.