Computational enzyme redesign: large jumps in function

Computational enzyme redesign allows large sequence jumps along complex and rugged protein-fitness landscapes, thus navigating to new functions in fitness landscapes with reduced experimental effort. Data-driven approaches are now offering new tools for discovery in numerous fields. Although their full potential remains to be realized, recent examples suggested that they can help to traverse fitness landscapes more efficiently. New machine-learning (ML) methods, such as deep-learning methods, have greatly promoted the demand for collection of more uniform and unbiased data sets of higher quality. Rising demands for enzymes in biotechnological applications have fueled efforts to tailor their properties towards desired functions, such as activity, selectivity, and stability. Computational methods are increasingly used in this task, providing designs that efficiently navigate large regions of sequence space with a greatly reduced experimental burden. With the improvement of enzyme redesign algorithms, model-based methods have achieved significant success in recent decades. Meanwhile, the rapid growth in protein databases has also promoted the development of data-driven approaches. Although data-driven approaches are just emerging, it will be exciting to see whether they can advance the field of enzyme redesign with the accumulation of more standard data, just as they are with structure prediction. Here, we present a brief overview of the field of computational enzyme redesign. We anticipate a marriage between model-based and data-based approaches which may offer opportunities to achieve more ambitious enzyme engineering goals in the coming years. Rising demands for enzymes in biotechnological applications have fueled efforts to tailor their properties towards desired functions, such as activity, selectivity, and stability. Computational methods are increasingly used in this task, providing designs that efficiently navigate large regions of sequence space with a greatly reduced experimental burden. With the improvement of enzyme redesign algorithms, model-based methods have achieved significant success in recent decades. Meanwhile, the rapid growth in protein databases has also promoted the development of data-driven approaches. Although data-driven approaches are just emerging, it will be exciting to see whether they can advance the field of enzyme redesign with the accumulation of more standard data, just as they are with structure prediction. Here, we present a brief overview of the field of computational enzyme redesign. We anticipate a marriage between model-based and data-based approaches which may offer opportunities to achieve more ambitious enzyme engineering goals in the coming years. an ML method based on multilevel neural network models, which can represent increasingly abstract concepts or patterns, level by level. aims to create artificial enzymes with desired functions that were not previously provided by nature. in sequence space determines the selection process of protein evolution. A protein fitness landscape describes how a given set of mutations affect the function of a protein of interest. incorporates the theozyme into large amount of natural protein folds and optimizes the surrounding residues to design artificial enzymes with specific functions. use an atomistic force field to describe the dynamic motions of macromolecules over time. uses geometric criteria (distances, angles, and dihedrals) to determine conformations that are close to the transition state of the reaction. use wave functions to describe the state of atoms and their fundamental particles. They can be used to predict the transition states of the desired reaction. employs QM calculations to determine an ideal geometrical arrangement of the active site capable of performing catalysis.