Reduction of monoclonal antibody viscosity using interpretable machine learning.

Early identification of antibody candidates with drug-like properties is essential for simplifying the development of safe and effective antibody therapeutics. For subcutaneous administration, it is important to identify candidates with low self-association to enable their formulation at high concentration while maintaining low viscosity, opalescence, and aggregation. Here, we report an interpretable machine learning model for predicting antibody (IgG1) variants with low viscosity using only the sequences of their variable (Fv) regions. Our model was trained on antibody viscosity data (>100 mg/mL mAb concentration) obtained at a common formulation pH (pH 5.2), and it identifies three key Fv features of antibodies linked to viscosity, namely their isoelectric points, hydrophobic patch sizes, and numbers of negatively charged patches. Of the three features, most predicted antibodies at risk for high viscosity, including antibodies with diverse antibody germlines in our study (79 mAbs) as well as clinical-stage IgG1s (94 mAbs), are those with low Fv isoelectric points (Fv pIs < 6.3). Our model identifies viscous antibodies with relatively high accuracy not only in our training and test sets, but also for previously reported data. Importantly, we show that the interpretable nature of the model enables the design of mutations that significantly reduce antibody viscosity, which we confirmed experimentally. We expect that this approach can be readily integrated into the drug development process to reduce the need for experimental viscosity screening and improve the identification of antibody candidates with drug-like properties.

Improved Diffusion Interaction Parameter Measurement to Predict the Viscosity of Concentrated mAb Solutions.

During the developability assessment of therapeutic monoclonal antibody (mAb) candidates, utilization of robust high-throughput predictive assays enables rapid selection of top candidates with low risks for late-stage development. Predicting the viscosities of highly concentrated mAbs using limited materials is an important aspect of developability assessment because high viscosity can complicate manufacturability, stability, and administration. Here, we report a high-throughput assay measuring protein-protein interactions to predict mAb viscosity. The diffusion interaction parameter (kD) measures colloidal self-association in dilute solutions and has been reported to be predictive of the mAb viscosity at high concentrations. However, kD of Amgen early stage IgG1 mAb candidates measured in 10 mM acetate at pH 5.2 containing sucrose and polysorbate (denoted A52SuT) shows only weak correlation to their viscosities at 140 mg/mL in A52SuT. We hypothesize that kD measured in A52SuT reflects primarily long-range electrostatic repulsions because most of these mAb candidates carry strong net positive charges in this low ionic strength formulation with pH (5.2) well below pI values of mAb candidates. However, the viscosities of high concentration mAbs depend heavily on short-range molecular interactions. We propose an improved kD method in which salt is added to suppress charge repulsions and to allow for detection of key short-range interactions in dilute solutions. Salt types and salt concentrations were screened, and an optimal salt condition was identified. This optimized method was further validated using two test mAb sets. Overall, the method improves the Pearson R2 between kD and viscosity (6-230 cP) from 0.24 to 0.80 for a data set consisting of 37 mAbs.