Single-cell dispensing and 'real-time' cell classification using convolutional neural networks for higher efficiency...

Area of Science:

• Biotechnology
• Machine Learning
• Cell Biology

Background:

• Single-cell dispensing is crucial for biopharmaceutical clonal cell line production.
• Automated cell isolation requires efficient methods for identifying viable single cells.

Purpose of the Study:

• To apply machine learning for real-time cell viability sorting during single-cell dispensing.
• To evaluate the performance of a shallow convolutional neural network (CNN) for cell image classification.
• To enhance clone recovery rates in cell line development.

Main Methods:

• Development and training of a shallow convolutional neural network (CNN) for classifying cell images.
• Validation of CNN performance using datasets from four different cell samples.
• Deployment of the trained CNN on a c.sight single-cell printer for real-time sorting.
• Assessment of clone recovery rates before and after implementing the AI-driven sorting system.

Main Results:

• An extremely shallow CNN outperformed more complex architectures for low-complexity cell image classification.
• The AI-driven sorting system predicted increased clone recovery across all investigated samples.
• On a sample with damaged cells, clone recovery improved from 27% to 73%.
• The frequency of dispensing viable cells increased by up to 65%, depending on the classification threshold.

Conclusions:

• Machine learning-based image classification enables effective real-time cell viability sorting.
• This approach significantly enhances the efficiency and speed of clonal cell line production.
• The versatile technology facilitates image-based cell sorting for various criteria using computer vision.