Machine learning‐assisted clone selection for intensified cell culture processes

Intensified fed-batch processes are becoming increasingly prevalent among biomanufacturers due to their superior space-time yields relative to traditional, non-intensified fed-batch processes. However, the shift towards intensified manufacturing has unexpectedly made optimal clone selection more challenging. Clone selection, traditionally an empirical screening of candidates on the basis of productivity, has been complicated by the observation that clonal productivity rankings can vary substantially between intensified and non-intensified processes. In the absence of a small-scale intensified empirical screen, clones destined for intensified manufacturing are selected based on their performance in the non-intensified process. This mismatch risks selection of suboptimal clones, representing a missed opportunity to generate additional product at no extra cost. This study presents a machine learning approach to improve clone selection for intensified processes by predicting and ranking clonal productivity. Models were trained on non-intensified clone performance data routinely collected during cell line development to facilitate seamless integration into existing capabilities. To evaluate generalizability and simulate real-world application, independent clone panels expressing two distinct monoclonal antibodies were withheld from model training and used in comparative testing against the legacy clone selection method. The ranking model identified the most productive clone of those expressing the first mAb, representing a 68.5% higher titer than the clone selected by the legacy method. For the second mAb, the ranking model identified the second most productive clone, resulting in a marginally lower titer. These results demonstrate the potential of machine learning models as practical tools for improving the selection of high productivity clones in intensified bioprocessing.