Developing a method for prediction of affinities for single and double protein mutants to different targets with similar structures.

Because of the obvious importance of serine proteases in numerous biological processes, and an abnormal regulation activity resulting in various types of cancer, many serine protease inhibitors have been designed in the past years, albeit with limited success. Indiscriminate inhibition of most serine proteases by such inhibitors inevitably leads to poorer outcomes. In the past years, it has become clear that inhibitors with narrow or single serine protease specificity hold much greater therapeutic potential; nevertheless, such specific inhibitors are difficult to obtain. Based on our published and preliminary data, we hypothesize that Amyloid-beta precursor protein inhibitor (APPI), a natural serine protease family inhibitor, can be engineered into highly selective serine protease inhibitor free of undesired off-target activities, to produce therapeutics targeting individual serine proteases with exquisite selectivity.

This study aims to provide a new approach for mapping the binding specificity landscapes of APPI, through a combination of experimental dual-target selective screening of mutant APPI library by yeast surface display (YSD), in silico next-generation sequencing and machine learning analysis, and a few experimental affinity data points on purified and YSD APPI protein variants. We show that our approach can generate binding affinity (and selectivity) values for numerous APPI mutants in complex with the human serine proteases Kallikrein- 6 (KLK6) and Mesotrypsin, and that the binding affinity (and selectivity) for these interactions could be quantified with high accuracy over a large range of affinity constants displayed by various APPI single and double mutants present, and even not existing in the library.