In Silico Screening of Computational Enzyme Designs

Computational enzyme design is a very promising area of research. However, current computational design tools suffer from a serious limitation: they cannot reliably predict active sequences and thus produce a large number of false positives. Consequently, investigators have had to rely on screening hundreds of designs to find a few active ones. In addition, both protein dynamics and explicit solvation are important aspects of enzyme function and are therefore critical to predicting enzymatic activity. Unfortunately, protein design methodology rarely takes these factors into account. Molecular dynamics (MD) simulations, on the other hand, can model both. In this work, we develop a MD protocol that provides a rei iable and computationally feasible way to pre-screen enzyme designs. Preliminary results have shown excellent agreement between experimental data and MD predictions. We tested the generality of the methodology on a larger and more diverse set of enzymes for which experimental data already exist. The method is general, sensitive (0.70) and specific (0.72). It also exhibits high negative predictive value (0.91) and promising computational performance, and wilt be useful for de novo computational enzyme design.