Life could not be sustained without the presence of enzymes, which are responsible for accelerating the chemical reactions in a biologically compatible timescale. Enzymes present other advantageous features such as high specificity and selectivity, plus they operate under very mild biological conditions. Inspired by these extraordinary characteristics, many scientists wondered about the possibility of designing new enzymes for industrially relevant targets. Unfortunately, none of the current enzyme design strategies can rapidly design tailor-made enzymes exhibiting high levels of activity at a reduced cost. This is limiting the general routine application of enzyme catalysis in industry, and thus the chemical manufacturing competitiveness.
Our goal is to develop a fast yet accurate computational enzyme design approaches for allowing the routine design of highly efficient enzymes. We combine computational chemistry, deep learning, graph theory, and computational geometry for controlling the complexity of enzyme catalysis and for developing a new computational pipeline to capture the chemical steps and conformational changes that take place along the enzyme catalytic cycle. Instead of relying on computationally expensive Molecular Dynamics (MD) simulations, we tuned the recently developed neural network AlphaFold2 (which is able to predict the three-dimensional structure of enzymes with high precision) for estimating the conformational flexibility of different enzyme variants.(1) Additionally, thanks to our new developed correlation-based methods focused on exploiting allostery operating in some enzymes,(2,3) we can predict mutations located at the enzyme active site (where the reaction happens), but also at positions far away from the reaction center.
