Antibodies are therapeutics derived from biological systems that are subject to non-natural stress throughout the manufacturing process. Therefore, antibodies are engineered to improve their biophysical properties to maximize recovery and yield. One antibody property of interest is thermostability. This is a difficult property to predict and has seen limited success with machine learning (ML) models because of the lack of data. We are addressing this problem by developing a phage display workflow that will allow us to train high-quality models and consequently improve our protein engineering suite.

• New phage display methodology to separate unstable from stable Fab sequences.
• New dataset containing hundreds of Fab sequences that are labelled unstable or stable at specific stress temperature.

We will train an ML model to predict the thermostability of Fab sequences. This will be incorporated into Abacus^TM to help assess the developability of potential therapeutics for our clients.

Abacus, our own suite of computational tools, leverages structure models, sequence information, and AI to guide and speed up biologic development.