CLANet: A comprehensive framework for cross-batch cell line identification using brightfield images.

Area of Science:

• Biomedical imaging
• Machine learning
• Cell biology

Background:

• Cell line authentication is critical for reproducible biomedical research.
• Supervised deep learning shows promise for cell line identification via imaging.
• Biological batch effects significantly challenge reliable cell line differentiation.

Purpose of the Study:

• To introduce CLANet, a framework for cross-batch cell line identification using brightfield images.
• To address challenges posed by biological batch effects in cell line identification.
• To improve the accuracy and reliability of cell line authentication across diverse experimental conditions.

Main Methods:

• CLANet utilizes brightfield images for cross-batch cell line identification.
• Employs a cell cluster-level selection for density variations and self-supervised learning for image quality.
• Incorporates multiple instance learning (MIL) with a time-series segment sampling module.

Main Results:

• CLANet effectively mitigates three distinct biological batch effects.
• Demonstrates superior performance compared to existing domain adaptation and MIL methods.
• Validated on 32 cell lines across 93 experimental batches from the AstraZeneca Global Cell Bank.

Conclusions:

• CLANet provides a robust solution for cell line identification across biological batches.
• The framework enhances reliability in cell line authentication despite variations in data generation.
• Paves the way for more accurate and dependable cell line identification in biomedical research.