Predicting Regulatory Motifs of Enzymes Using Machine Learning

Abstract

Enzymes perform critical biochemical reactions in the cell, but they are not given a free ride to roam in the cellular milieu. Specifically, a variety of regulatory factors such as substrate, product, inducer, or activator interact with different domains and motifs on the enzyme to enact regulatory action that tunes, in broad strokes, enzyme activity. Much of such understanding comes from detailed biochemical assays augmented with insights from structural biology. Progress in elucidation of regulatory actions on enzymes is slow and defined by generalizing characterized phenomena into broad categories. But, much remains unknown. Recent advances in fast computational protein structure prediction by machine learning tools generates a larger set of structures for which structural biologists could link amino acid sequence to enzyme structure, and therefore predict putative binding sites for regulatory factors on proteins. Such insights could be further augmented by convolution neural network (CNN) exploration of protein-ligand binding affinity that annotated particular structural folds from structure prediction with biological function. While nascent, the combination of the above machine learning tools could revolutionize the speed and accuracy at which regulatory motifs on new enzymes could be predicted, which narrows the search space for biochemical experimental verification.