Deconvolving multiplexed protease signatures with substrate reduction and activity clustering

Proteases are pleiotropic, promiscuous enzymes that degrade proteins and peptides, which drive important processes in health and disease. The ability to quantify the activity of protease signatures by sampling with Massively Multiplexed Activity (MMA) libraries will provide unparalleled biological information. Under such a framework, a designed library of peptide substrates is exposed to a cocktail of proteases, the cleavage velocity of each substrate is measured, and individual protease activity levels are inferred from the data. Previous studies have developed individual protease sensors, but multiplexed substrate cleavage data becomes difficult to interpret as the number of cross-cutting proteases increases. Computational methods for parsing this data to estimate individual protease activities primarily use an extensive compendium of all possible protease-substrate combinations, which require impractical amounts of training data when scaling up to MMA libraries. Here we provide a computational method for estimating protease activities efficiently by reducing the number of substrates and clustering proteases with similar cleavage activities into families. This method is scalable and will enable the future use of MMA libraries with applications spanning therapeutic and diagnostic biotechnology.