SubTuner leverages physics-based modeling to complement AI in enzyme engineering toward non-native substrates

We developed SubTuner, a physics-based computational tool that tackles the challenge of identifying enzyme mutants with enhanced activity for specified non-native substrates. To test the performance of SubTuner, we designed three tasks - all aiming to identify beneficial anion methyltransferase mutants for synthesis of non-native S-adenosyl-l-methionine analogs: first in the conversion of ethyl iodide from a pool of 190 AtHOL1 single-point mutants for an initial test of accuracy and speed; second of ethyl, n-propyl, cyclopropylmethyl, and phenethyl iodide from a pool of 600 acl-MT multi-point mutants for a test of generalizability; and eventually of bulkier substrates for AtHOL1 combined with experimental characterization for a test of a priori predictivity. All tests demonstrated SubTuner's ability to accelerate enzyme engineering for non-native substrates, superior to existing bioinformatics and machine learning-based tools. SubTuner, with its physical hypothesis, quantitative accuracy, and mechanism-informing ability, holds a significant potential to aid enzyme engineering for substrate scope expansion.