Rapid Estimation of Size-Based Heterogeneity in Monoclonal Antibodies by Machine Learning-Enhanced Dynamic Light Scattering

Aggregation of monoclonal antibody therapeutics is a serious concern that is believed to impact product safety and efficacy. There is a need for analytical approaches that enable rapid estimation of mAb aggregates. Dynamic light scattering (DLS) is a well-established technique for estimating the average size of protein aggregates or for evaluating sample stability. It is usually used to measure the size and size distribution over a wide range of nano- to micro-sized particles using time-dependent fluctuations in the intensity of scattered light arising from the Brownian motion of particles. In this study, we present a novel DLS-based approach that allows us to quantify the relative percentage of multimers (monomer, dimer, trimer, and tetramer) in a monoclonal antibody (mAb) therapeutic product. The proposed approach uses a machine learning (ML) algorithm and regression to model the system and predict the amount of relevant species such as monomer, dimer, trimer, and tetramer of a mAb in the size range of 10-100 nm. The proposed DLS-ML technique compares favorably to all potential alternatives with respect to the key method attributes, including per sample cost of analysis, per sample time of data acquisition along with ML-based aggregate prediction (<2 min), sample requirements (<3 μg), and user-friendliness of analysis. The proposed rapid method can serve as an orthogonal tool to size exclusion chromatography, which is the current industry workhorse for aggregate assessment.