Towards image-based cancer cell lines authentication using deep neural networks.

Area of Science:

• Biotechnology
• Bioinformatics
• Computational Biology

Background:

• Short tandem repeat (STR) analysis is a reliable method for cell line genetic origin determination but is costly and time-consuming.
• Current methods cannot distinguish between isogenic cell lines (e.g., clonal sublines or drug-adapted variants).
• There is a need for low-cost, low-effort methods for routine cell line identity monitoring and isogenic cell line authentication.

Purpose of the Study:

• To automate cell line identification and authentication using image-based analysis and deep convolutional neural networks.
• To develop a complementary method to STR analysis for routine laboratory use.
• To enable the discrimination of isogenic cell lines, including drug-adapted sublines.

Main Methods:

• Two deep convolutional neural network models, MobileNet and InceptionResNet V2, were trained.
• Models were trained to classify eight classes: four parental cancer cell lines and their drug-adapted sublines (cisplatin and oxaliplatin).
• Performance was evaluated using tenfold cross-validation, F1-score, and area under the curve (AUC).

Main Results:

• The InceptionResNet V2 model achieved an average F1-score of 0.91 and an AUC of 0.95 for the eight-class problem.
• The best model demonstrated high performance in classifying parental cell lines (average F1-score of 0.94) and drug-adapted cells (average F1-score of 0.96) separately.
• This study serves as a proof of principle for using image-based deep learning for cancer cell line authentication.

Conclusions:

• Deep learning-based image analysis can automate cell line identification and authentication.
• This approach offers a potential low-cost, low-effort complement to traditional methods like STR analysis.
• The developed methodology can facilitate routine monitoring of cell line identity, including isogenic variants.