Accurate computational evolution of proteins and its dependence on deep learning

Abstract

Enzyme is the major workhorse to carry out the diverse cellular functions. It catalyzes the biological reactions with a high specificity, with its topology playing a crucial role. For ecologically safe production of numerous bioproducts including drugs and chemicals, we have been striving to design the industrially useful enzyme molecules with highly improved catalytic capability. As the sequence space is enormous for an enzyme, its quick and effective exploration is quite improbable for the mutagenesis studies whose accuracy is greatly reliant on the prior information of the mutated sites and the extent of rigorous screening of the mutant libraries. Although directed evolution methods significantly aid the construction of a functionally improved molecule, their credibility depends on the successful excavation of the functionally similar sequence space in the available databases, encompassing billions of proteins. As deep learning methods aid us to extensively uncover the underlying network of all the key catalytic positions without any experimental data, their implementation has reliably increased the accuracy of directed evolution. The chapter comprehensively explains data mining and deep learning methods to further showcase their importance in enzyme engineering methods. The key biological and algorithmic limitations of these deep learning methodologies are lastly highlighted.