Learning How to Make Mini-Proteins that Bind to Specific Target Proteins

Abstract

Antibodies seem to have no trouble targeting and binding to specific natural proteins. But how do they do it? Human attempts to copy this ability often directly copy antibodies; either by borrowing their loops or their hotspots. While humans can analyze interactions between antibodies and their targets or interactions between native protein complexes, the proof of understanding requires one to make and test predictions. In this work, we set out to learn the principles that allow for specific protein binding. Leveraging the modern biochemist’s toolkit, we tested more than one million de novo mini-protein binders (< 65aa) to more than twenty natural targets. With the goal of learning, not copying, we threw out the native complex information and discovered our own hotspots and interactions. Some of the target sites did not even have natural partners. By iterating on design-test-learn, we were able to uncover secrets about protein folding and protein binding. We found that the hydrophobic effect and hydrophobic packing dominate the correlations. While future work may find ways to bind to polar patches, for now, we have a sure-fire way to make binders to hydrophobic patches.