Inspiring Innovation in Formulation, Bioprocessing and Drug Delivery

Keynote: Predict and Overcome Developability Challenges in High-Concentration Formulations Through Machine Learning

Keynote Speaker(s)

• 

  Pin-Kuang Lai, PhD (he/him/his)

  Assistant Professor
  Stevens Institute of Technology
  Warren, New Jersey

  Handouts

Moderator(s)

• 

  Yongchao Su, PhD

  Senior Principal Scientist
  Merck & Co, Inc.
  Rahway, New Jersey

High-concentration antibody stability is a critical factor in the development of subcutaneous injections, with key challenges including viscosity, aggregation, and bioavailability. However, early-stage drug development often struggles with limited material availability for high-concentration measurements. Although machine learning has shown promise in predicting these properties, its progress has been hindered by the scarcity of comprehensive datasets.
To address this, teams at Stevens and AstraZeneca collaborated to generate an extensive dataset of 229 high-concentration antibodies with viscosity measurements. This effort led to the development of DeepViscosity, an ensemble deep learning model capable of classifying monoclonal antibodies with low viscosity (≤20 cP) and high viscosity (>20 cP) at 150 mg/mL using sequence data alone. When tested on two independent datasets of 16 and 38 mAbs, DeepViscosity demonstrated impressive accuracy, achieving 87.5% and 89.5%, respectively, and outperforming other predictive methods.
Additionally, the Stevens team developed bioavailability prediction tools leveraging protein language models, achieving prediction accuracies approaching 90%. These advancements mark significant strides in using machine learning to overcome challenges in high-concentration antibody formulation and subcutaneous administration development.