High-throughput discovery of the determinants of stress-induced aggregation in miniproteins for hyper-stable therapeutic design

We propose an approach to learn how to design aggregation-resistant protein therapeutics. Therapeutic proteins engineered to be aggregation-resistant offer several advantages for drug delivery, such as: (1) decreased risk of forming immunogenic aggregates, (2) ability to be formulated at higher concentrations or with other biologics, and (3) less reliance on the cold chain for transportation and storage. We will measure aggregation resistance after thermal stress exposure for thousands of proteins at one time. We will use machine learning analysis to predict aggregation based on protein sequence and structural features. Then we will robustly test our model by designing thousands of new proteins predicted to more aggregation resistant and testing them experimentally. We will learn how to improve the model by comparing the experimental and predicted aggregation resistance. Using this iterative approach, we will learn how to design proteins with extremely high aggregation resistance.