PROPERMAB: an integrative framework for in silico prediction of antibody developability using machine learning.

Area of Science:

• Biopharmaceutical development
• Computational biology
• Protein engineering

Background:

• Therapeutic molecule selection prioritizes efficacy and safety, with developability assessed late.
• Late-stage developability assessment delays drug development and increases costs.
• Early evaluation of antibody developability is crucial for efficient drug discovery.

Purpose of the Study:

• To develop a computational framework for early, large-scale prediction of monoclonal antibody developability properties.
• To enable cost-effective and high-throughput assessment of antibody developability using machine learning.
• To accelerate the discovery-to-clinic timeline for antibody therapeutics.

Main Methods:

• Developed PROPERMAB (PROPERTIES of Monoclonal AntiBodies), a computational framework for in silico prediction.
• Utilized custom molecular features and machine learning modeling for property prediction.
• Pre-trained models to predict structure-derived features directly from antibody sequences.

Main Results:

• Successfully developed models to predict antibody hydrophobic interaction chromatography retention time and high-concentration viscosity.
• Demonstrated rapid and accurate prediction of structure-derived features from sequences.
• Showcased the scalability of the approach for repertoire-scale sequence datasets.

Conclusions:

• PROPERMAB enables efficient in silico prediction of antibody developability properties.
• Early prediction of developability using sequence-derived features accelerates biopharmaceutical research.
• Computational approaches offer a powerful alternative to experimental methods for antibody developability assessment.