High-throughput developability assays enable library-scale identification of producible protein scaffold variants
Alexander W. Golinski, Katelynn M. Mischler, Sidharth Laxminarayan, Nicole Neurock, Matthew Fossing, Hannah Pichman, Stefano Martiniani, Benjamin J. Hackel
Abstract
variants was calculated via deep sequencing of populations sorted by proxied developability. We identified the most informative assay combination via cross-validation accuracy and correlation feature selection and demonstrated the ability of machine learning models to exploit nonlinear mutual information to increase the assays' predictive utility. We trained a random forest model that predicts expression from assay performance that is 35% closer to the experimental variance and trains 80% more efficiently than a model predicting from sequence information alone. Utilizing the predicted expression, we performed a site-wise analysis and predicted mutations consistent with enhanced developability. The validated assays offer the ability to identify developable proteins at unprecedented scales, reducing the bottleneck of protein commercialization.