Predicting expression levels of de novo protein designs in yeast through machine learning.

Abstract

Protein engineering has unlocked potential for the biomolecular synthesis of amino acid sequences with specified functions. With the many implications in enhancing therapy development, developing accurate functional models of recombinant genetic outcomes has shown great promise in unlocking the potential of novel protein structures. However, the complicated nature of structural determination requires an iterative process of design and testing among an enormous search space for potential designs. Expressed protein sequences through biological platforms are necessary to produce and analyze the functional efficacy of novel designs. In this study, machine learning principles are applied to a yeast surface display protocol with the goal of optimizing the search space of amino acid sequences that will express in yeast transformations. Machine learning networks are compared in predicting expression levels and generating similar sequences to find those likely to be successfully expressed.