Data-Driven Antibody Formulation in a Self-Driving Lab

Project Summary This project aims to develop the foundational data science methods and pipeline for efficient and exceptional monoclonal antibody (mAb) formulation. Current formulation efforts by pharmaceutical companies and contract research organizations use a combination of rational design, design of experiments (DOE), and high throughput screening to identify Generally Recognized as Safe (GRAS) additive combinations for stable formulation. However, as the demand for subcutaneous dosing increases, so does the need for higher performance outcomes with high concentrations, low viscosity, and ambient temperature stability. Unfortunately, these rare properties are difficult to discover due to the vast and highly dimensional formulation parameter space. Therefore, we propose that data science methods such as artificial intelligence / machine learning (AI/ML) are ideally suited to drive these formulation campaigns. Also, their strategic integration with advanced automation and analytics provides a unique opportunity to develop a self-driving bioformulation laboratory which would greatly increase formulation efficiencies. To build towards this overarching vision, Aim 1 will prototype experimental and data handling methodologies, Aim 2 will perform a multi-objective AI/ML-driven formulation search, and Aim 3 will synergize these methods into one system for autonomous quantitative structure-activity relationship (QSAR) modeling. Efficiency gains from data driven bioformulation on a robotic platform will allow drug developers to test the formulation potential of their drugs at much earlier stages of drug development. In the long-term, this may enable remarkably stable dosing forms for at home subcutaneous administration and ambient temperature storage.