High-concentration monoclonal antibody (mAb) formulations are essential for subcutaneous drug delivery, but elevated viscosity—driven by complex and poorly understood protein-protein interactions (PPIs)—poses challenges to manufacturability, syringeability, and injectability. Existing predictive models often lack sufficient data for robust validation, and experimentally measuring viscosity at high concentrations is both costly and resource intensive. There is a pressing need for more efficient approaches to assess viscosity using computational models or high-throughput experimental techniques with minimal sample requirements. We developed DeepViscosity, an ensemble model comprising 102 artificial neural networks trained on viscosity data from 229 high-concentration mAbs—the largest dataset to date in this field. The model classifies mAbs as low-viscosity (≤ 20 cP) or high-viscosity (> 20 cP) at 150 mg/mL based on sequence-derived features, achieving up to 89.5% accuracy on independent test sets. In parallel, we explore the utility of Small-Angle X-ray Scattering (SAXS) as a high-throughput method to investigate PPIs in solution. SAXS captures changes in scattering profiles that reflect repulsive or attractive interparticle interactions across a range of concentrations, providing insight into self-association and crowding effects. We propose a SAXS-based strategy for early identification of low-viscosity mAbs at high concentrations. Notably, a SAXS-derived interparticle structure factor parameter measured at 10 mg/mL showed strong correlation with viscosity measured at 150 mg/mL, outperforming traditional diffusion interaction (kD) metrics. Together, DeepViscosity and SAXS offer complementary strategies to accelerate the development of low-viscosity mAbs, reduce formulation risks, and streamline early-stage candidate selection.
