Deep learning-based classification of breast cancer cells using transmembrane receptor dynamics

Area of Science:

• Biophysics
• Cancer Biology
• Machine Learning

Background:

• Transmembrane receptor motions on cancer cells offer insights into cell phenotypes.
• Conventional analysis methods like mean-squared displacement plots can lose critical trajectory information.
• Characterizing cancer cell types and phenotypes is crucial for effective treatment strategies.

Purpose of the Study:

• To employ deep learning for classifying breast cancer cell types based on epidermal growth factor receptor (EGFR) motion trajectories.
• To develop an alternative method for cancer cell characterization using dynamic, biophysical features accessible from receptor movements.

Main Methods:

• An artificial neural network was developed and trained on EGFR motion trajectories from six distinct breast cancer cell lines.
• The model was evaluated on its ability to classify cell lines and predict receptor status.
• Epithelial-mesenchymal transition (EMT) was induced in specific cell lines to further validate the deep learning model's predictive capabilities.

Main Results:

• The deep learning model achieved 83% accuracy in classifying trajectories within individual cell lines.
• The model demonstrated 85% accuracy in predicting cancer cell receptor status.
• Following EMT induction, benign and noninvasive cancer cells showed increased classification as triple-negative (TN) cells, validating the model's sensitivity to phenotypic changes.

Conclusions:

• Deep learning analysis of EGFR trajectories provides a powerful, alternative method for breast cancer cell classification.
• This approach leverages dynamic, biophysical cell properties, complementing traditional image-based methods.
• The study demonstrates the potential of using receptor dynamics for understanding and classifying cancer cell phenotypes.