Deep learning-based in silico labeling for analyzing morphological features of MSCs to predict immunomodulatory capacity.

Cellular morphology, a critical manifestation of biological characteristics, is linked to functions. In traditional cell detection, invasive labeling and detection methods not only compromise cellular viability but also entail labor-intensive workflows. Here we presented a non-invasive artificial intelligence framework that integrated deep learning (DL) and machine learning (ML) to predict the immunomodulatory capacity of mesenchymal stem cells (MSCs) through morphological profiling. The improved PreAct-ResNet50 encoder-decoder architecture was used to achieve high-accuracy instance segmentation of cells and nuclei, enabling quantification of morphological features. A LightGBM-based predictive model was subsequently employed to predict MSCs immunomodulatory biomarkers through morphological features. This dual-model system demonstrated satisfactory cell segmentation and biological characteristics prediction capabilities through performance testing. Our method provided an efficient, non- invasive tool for real-time MSCs potency assessment, which could enhance quality controls in cell therapy manufacturing.