In Silico Screening of Computational Enzyme Designs

BERNARDO SOSA PADILLA ARAUJO; THOMAS MILLER; STEPHEN MAYO

San Diego, California

Conferencia; The 26th Annual Symposium of The Protein Society; 2012

The Protein Society

Computational enzyme design is a very promising areaof research. However, current computational designtools suffer from a serious limitation: they cannotreliably predict active sequences and thus producea large number of false positives. Consequently,investigators have had to rely on screening hundredsof designs to find a few active ones. In addition, bothprotein dynamics and explicit solvation are importantaspects of enzyme function and are therefore critical topredicting enzymatic activity. Unfortunately, proteindesign methodology rarely takes these factors intoaccount. Molecular dynamics (MD) simulations, on theother hand, can model both. In this work, we develop aMD protocol that provides a reliable and computationallyfeasible way to pre-screen enzyme designs. Preliminaryresults have shown excellent agreement betweenexperimental data and MD predictions. We tested thegenerality of the methodology on a larger and morediverse set of enzymes for which experimental dataalready exist. The method is general, sensitive (0.70) andspecific (0.72). It also exhibits high negative predictivevalue (0.91) and promising computational performance,and will be useful for de novo computational enzymedesign. This work was supported by a DARPA ProteinDesign Processes grant and a National Security Scienceand Engineering Faculty Fellowship awarded to S.L.M.