Saxs-guided coarse-grained model development to elucidate antibody-antibody interactions for subcutaneous drug delivery

The National Institute of Allergy and Infectious Diseases awarded The Trustees Of The Stevens Institute Of Technology $192,765 on September 1, 2026, under the Allergy and Infectious Diseases Research program (CFDA 93.855).

The award funds development of a machine learning-guided coarse-grained modeling framework to predict antibody behavior in high-concentration formulations for subcutaneous drug delivery. The research integrates small-angle X-ray scattering (SAXS), viscosity, and aggregation data to parameterize protein-protein interactions across diverse therapeutic antibodies. The objective is to establish a predictive platform connecting antibody sequence to macroscopic solution properties, enabling rational design of low-viscosity, stable formulations that reduce injection difficulty and aggregation risk in subcutaneous administration. The work addresses a key challenge in formulation science: high viscosity and irreversible aggregation currently limit clinical feasibility and manufacturability of concentrated antibody therapeutics.

Performance occurs in Hoboken, New Jersey, with a period of performance running through August 31, 2028. The assistance type is a Project Grant, classified as R21 funding under the NIH structure.

Abstract

High-concentration antibody formulations are critical for subcutaneous (SC) administration, offering improved patient adherence, reduced healthcare burden, and convenient self-administration. However, elevated viscosity and irreversible aggregation present major challenges, limiting clinical feasibility and manufacturability. High viscosity hinders injection through fine-gauge needles, while aggregation reduces potency, shelf life, and safety. Although numerous formulation strategies have been explored, most remain empirical due to limited understanding of how antibody sequence, structure, and environmental factors influence viscosity and aggregation. Computational models have been developed to study protein-protein interactions (ppis), yet their predictive accuracy remains restricted by simplified parameterization and limited experimental validation. This limitation hampers rational design of low-viscosity, stable antibody formulations. The long-term goal of this research is to establish a predictive modeling framework that connects antibody sequence to macroscopic solution behavior. The overall objective is to develop a machine learning (ML)-guided coarse-grained (CG) modeling framework that integrates high-throughput small-angle x-RAY scattering (saxs), viscosity, and aggregation data to parameterize ppis across diverse antibodies. Our central hypothesis is that ML-refined CG parameters can accurately capture mesoscale structure factors and macroscopic viscosity and aggregation, providing a unified, sequence-based prediction platform. This framework will bridge molecular interactions with bulk solution properties and enable data-driven formulation design. To test this hypothesis, we will pursue three specific aims. Aim 1: perform high-throughput saxs measurements for ~230 therapeutic antibodies under varying conditions and correlate s(q) with viscosity and aggregation profiles. Aim 2: develop saxs-guided CG- ML models integrating hydrodynamic viscosity calculations and aggregation analysis from cluster formation rates, with ML models predicting CG parameters from sequence descriptors (charge, hydropathy, deepsp, and protein language embeddings). Aim 3: evaluate model generalizability for bispecific antibodies (bsabs) and bispecific t-cell engagers (bites) to demonstrate transferability and robustness. This study is highly innovative in integrating saxs, viscosity, and aggregation data at scale to parameterize and validate CG-ML models, providing the first quantitative link between molecular-scale interactions and macroscopic rheological behavior. The collaborative framework with astrazeneca and brookhaven national laboratory enables unprecedented experimental throughput and data diversity. The successful completion of this project will deliver a generalizable computational-experimental platform for predicting, optimizing, and accelerating the development of low- viscosity, stable antibody therapeutics, ultimately advancing next-generation subcutaneous biologics.