Poster Session - Abstract # 8

Can We Identify Enzyme Active Sites Without Experimentation?

Ryan Feehan¹ and Joanna S.G. Slusky^1,2

¹The Center for Computational Biology, The University of Kansas, Lawrence, KS, USA; ²Department of Molecular Biosciences, The University of Kansas, Lawrence, KS, USA

Enzymes account for over half of solved protein structures and are known to play a role in over 8,000 reactions.  Despite the prevalence and importance of enzymes, the ability to identify enzyme active sites remains an unsolved, challenging task.  Our lab combined structure-based machine learning, metalloproteins, and physicochemical properties to distinguish between very similar sites, metalloenzyme active sites and inactive metal binding sites.  With a 92% precision and 94% recall, our model outperforms alternative enzyme prediction methods that utilize evolutionary information and large sequence datasets.  Moreover, on a set of predicted structures for proteins with no solved crystal structure, our model achieved 90-97% accuracy depending on the quality of the predicted structures.  Finally, to better understand the principles governing enzymatic activity, regardless of reaction type, we analyzed which physicochemical properties were fundamental for our model’s success.  We anticipate that our model’s ability to correctly identify catalytic metal sites could enable identification of new enzymatic mechanisms and improve de novo enzyme design success rates.