Machine learning and molecular modeling based design of nanobodies targeting human serotonin transporter and receptor.

Abstract

Design of nanobodies have emerged as a new trend in antibody engineering, leveraging their unique properties including high stability, solubility, and the ability to bind to challenging targets such as membrane proteins. The application of computational strategies is pivotal for refining the efficacy of protein binders like nanobodies by broadening the sequence diversity, forecasting and bolstering their binding potency, selectivity, and overall performance. Recent advancements in computational techniques, such as machine learning algorithms and physics-based molecular modeling have significantly improved the design and development of nanobodies. These techniques allow for the precise modeling of nanobody-target interactions, enabling the identification of key residues responsible for binding and the prediction of potential conformational changes. In this study, five parental nanobodies binding to GPCRs and transporters were first used as template to create in silico nanobody libraries with the SCHEMA algorithm. Then, their binding potential and function to GPCRs or transporters were predicted by pre-trained machine learning models. The sequences above a threshold were processed with Rosetta and AlphaFold2 for 3D structural predictions. To further identify optimal conformations of specific nanobodies theoretically binding to 5-HT1AR or SERT, protein-protein docking by RosettaDock were performed. Finally, based on these model complexes, new nanobodies were redesigned, resulting in 21 and 18 candidates with enhanced binding to 5-HT1AR and SERT, respectively.