High-Throughput Computational Screening of Solid-Binding Peptides

Inspired by biomineralization, a naturally occurring, protein-facilitated process, solid-binding peptides (SBPs) have gained much attention for their potential to fabricate various shaped nanocrystals and hierarchical nanostructures. The advantage of SBPs over other traditionally used synthetic polymers or short ligands is their tunable interaction with the solid material surface via carefully programmed sequence and being solution-dependent simultaneously. However, designing a sequence with targeted binding affinity or selectivity often involves intensive processes such as phage display, and only a limited number of sequences can be identified. Other computational efforts have also been introduced, but the validation process remains prohibitively expensive once a suitable sequence has been identified. In this paper, we present a new model to rapidly estimate the binding free energy of any given sequence to a solid surface. We show how the overall binding of a polypeptide can be estimated from the free energy contribution of each residue based on the statistics of the thermodynamically stable structure ensemble. We validated our model using five silica-binding peptides of different binding affinities and lengths and showed that the model is accurate and robust across a wider range of chemistries and binding strengths. The computational cost of this method can be as low as 3% of the commonly used enhanced sampling scheme for similar studies and has a great potential to be used in high-throughput algorithms to obtain larger training data sets for machine learning SBP screening.