Machine learning for early clone screening in cell line development via automated image analysis | Poster Board #142

Stable clonal cell lines are essential for the production of biologic therapeutics. Cell line development (CLD) is often labor-intensive, requiring the screening of thousands of clones to identify those with high productivity. The complexity of certain modalities further necessitates customized screens and the evaluation of more variants.

To address these challenges, we explored the use of machine learning (ML) via image analysis of early colony growth, aiming to automate, streamline, and accelerate the clone screening process. Using a high-throughput imaging platform, we collected image data sets that captured features during early clone outgrowth. We then trained and tested ML models using primary titer screening data from multiple assets to predict the most promising clones for secondary clone screening.

Our automated process generates model scores, significantly reducing bench time and improving overall throughput. This ML-driven imaging analysis workflow shortens CLD timelines by 3 weeks without compromising the ability to identify high producers. Additionally, it allows for the confident expansion of fewer clones, thereby reducing resource expenditure. We anticipate that further optimization of our models, based on production study data, will enhance predictive capabilities and generalize across various complex biologics, ultimately improving therapeutic development.