(169db) Virtual High Throughput Screening of Car9 Isomers to Study Sequence-Binding Relationships

The design of new solid-binding peptides (SBPs) is a non-trivial task driven by the relationship between the SBP sequence and the binding energy. This relationship is not deterministic from the amino acid (AA) sequence alone. This can be illustrated by Car9 (DSARGFKKPGKR). According to previous studies, with five positively charged AAs, Car9 binds strongly to a quartz surface (with –OH^- ions). However, not all of the five positively charged AAs participate in binding. The same has been found for a few selected mutants or sequence isomers of Car9. Therefore, we use Car9 and its ~20 million possible sequence isomers to study sequence-binding relationships. We narrow down this search space using unsupervised and semi-supervised clustering algorithms to end up with 1000 clusters with centers corresponding to 1000 peptides, which form a reasonable representation of the Car9 isomer space. We rank the 1000 resulting peptide sequences based on their sequence ‘similarity’ (calculated using metrics like Levenshtein and hamming distances) with Car9. We perform high throughput computational screening of the 1000 sequences using a new method of rapidly estimating binding energies developed by our group. We then determine if sequences ‘similar’ to Car9 have similar binding energies to the quartz surface. This would be a starting point for creating design rules for SBPs based on physics-agnostic similarity metrics. Understanding these relationships between the AA sequence and binding behavior is a step toward the inverse design of SBPs.