Associate Director, Early-Stage Formulation Sciences, Biopharmaceutical Development AstraZeneca Clarksburg, Maryland
Machine learning has been recently applied to predict antibody aggregation rates and viscosity. In this study, we measured accelerated aggregation rates at 45°C and viscosity at 150 mg/ml for 20 clinical stage antibodies. Using features calculated from intrinsic sequences and molecular dynamics simulations, we have applied machine learning to develop a k-nearest neighbors regression model with two features- spatial positive charge map (CDRH2) and solvent accessible surface area of hydrophobic residues- to predict aggregation rates and a logistic regression classification model with two features- spatial negative charge map (HC-Fv) and spatial negative charge map (LC-Fv) to predict viscosity. The aggregation rates and viscosity models can be applied to predict early to clinical stage antibody stability to facilitate pharmaceutical development.
Learning Objectives:
Discuss challeges in high concentration therapeutic protein formulation development.
Discuss the application of an in silico apprach to guide early developability.
Understand how machine learning methods combined with sequences and structural features obtained from MD simulations may be used to find predictive models for antibody aggregation and viscosity at high concentrations.
Define computational homology modeling tools and machine learning methods that may be used in R&D workstreams to identify potential liabilities and accelerate development activities.