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

Area of Science:

• Biotechnology
• Cell Biology
• Artificial Intelligence

Background:

• Cellular morphology is crucial for biological functions.
• Traditional cell detection methods are invasive and labor-intensive.
• Assessing mesenchymal stem cells (MSCs) potency is vital for cell therapy.

Purpose of the Study:

• To develop a non-invasive artificial intelligence (AI) framework for predicting MSC immunomodulatory capacity.
• To utilize morphological profiling for real-time MSC potency assessment.
• To enhance quality control in cell therapy manufacturing.

Main Methods:

• Integrated deep learning (DL) and machine learning (ML) models.
• Employed an improved PreAct-ResNet50 encoder-decoder for cell and nuclei instance segmentation.
• Utilized a LightGBM model for predicting immunomodulatory biomarkers from morphological features.

Main Results:

• Achieved high-accuracy instance segmentation of cells and nuclei.
• Successfully predicted MSC immunomodulatory biomarkers using morphological features.
• Demonstrated satisfactory performance in cell segmentation and biological characteristic prediction.

Conclusions:

• The AI framework offers an efficient, non-invasive tool for real-time MSC potency assessment.
• This approach can significantly improve quality control in cell therapy manufacturing.
• Morphological profiling via AI provides a viable alternative to traditional methods.