Early identification of high-performing biopharmaceutical CHO cell lines using label-free multimodal optical microscopy and machine learning
This is the code repository for the paper: [DOI will be updated].
The selection of high-performing cell lines is a crucial step in the biopharmaceutical industry to ensure the development of molecules meets quality and manufacturability criteria. However, this process is often time- and labor-intensive, involving screening thousands of clones over several months. To address this challenge, we investigated the capability of label-free multimodal optical microscopy, which enables non-perturbative profiling of biological samples, offering rich structural and metabolic information. Specifically, we employed simultaneous label-free autofluorescence multiharmonic microscopy and fluorescence lifetime imaging microscopy to characterize different biopharmaceutical Chinese hamster ovary (CHO) cell lines in the early stages (i.e., passage 0 to passage 2). To perceive the rich information and enable accurate classification of individual cells, a machine learning-assisted single-cell analysis pipeline was developed. Remarkably, the ML classifiers achieved average balanced accuracies exceeding 95% for monoclonal cell lines as early as passage 2 using 10-fold Monte Carlo cross-validation. Analysis of feature importance revealed that correlation and co-occurrence features played a pivotal role in classifying cell lines across passages, underscoring the importance of multimodal imaging. By integrating optical bioimaging and ML techniques, this study offers a promising solution to expedite the cell line selection process, reducing time and resources while ensuring the identification of high-performance biopharmaceutical cell lines.