AutoGater: a weakly supervised neural network model to gate cells in flow cytometric analyses.

Area of Science:

• * Biotechnology and Biomedical Engineering
• * Cell Biology and Immunology

Background:

• * Flow cytometry is a powerful technique for cell analysis, but dead cells can significantly interfere with results.
• * Current methods for excluding dead cells often rely on fluorescent stains, which add time and consume valuable fluorescence channels.
• * There is a need for efficient, stain-free methods to accurately identify viable cells in flow cytometry data.

Purpose of the Study:

• * To develop and validate a stain-free deep learning approach for identifying and excluding dead cells in flow cytometry.
• * To demonstrate the efficacy of AutoGater in separating healthy cells from dying and dead populations.

Main Methods:

• * Development of AutoGater, a weakly supervised deep learning model.
• * Utilizing only light-scatter channels for cell population discrimination.
• * Validation against traditional methods involving fluorescent stains (e.g., Sytox) and Colony Forming Units (CFUs).

Main Results:

• * AutoGater successfully separates healthy cell populations from unhealthy and dead cells using only light-scatter data.
• * The model provides stain-free identification of viable cells, preserving fluorescence channels for other markers.
• * AutoGater harmonizes measurements of cell death, correlating with traditional viability assays.

Conclusions:

• * AutoGater offers a robust and efficient stain-free alternative for dead cell exclusion in flow cytometry.
• * This method simplifies experimental workflows and enhances the capacity for multi-parameter analysis.
• * The deep learning approach provides a reliable way to improve the accuracy of flow cytometry data interpretation.