Machine Learning Designed for Any Hematologic Flow Cytometry Data Set.

Area of Science:

• Hematology
• Computational Biology
• Bioinformatics

Background:

• Flow cytometry is crucial for single-cell protein analysis in hematology.
• Manual interpretation of flow cytometry data is time-consuming and prone to interrater variability.
• Existing automated tools often lack flexibility, requiring specific diagnostic setups and limiting widespread adoption.

Purpose of the Study:

• To develop a versatile software package and web app, diagnFlow, for automated analysis of diverse clinical flow cytometry datasets.
• To demonstrate the clinical utility and benefits of diagnFlow, particularly in lymphoma diagnosis.
• To provide a data set-agnostic solution for flow cytometry analysis.

Main Methods:

• Development of the diagnFlow software package and accompanying web application.
• Creation of automated analysis workflows using diagnFlow for specific diagnostic tasks.
• Evaluation of diagnFlow's performance against manual interpretation and other automated methods.
• Validation of the web app version on independent datasets.

Main Results:

• Automated workflows developed with diagnFlow consistently outperformed manual interpretation.
• An interpretable and efficient workflow was identified and deployed as a user-friendly web app.
• The diagnFlow approach demonstrated superior performance compared to a leading clustering-based method.
• Wet laboratory data confirmed the biological basis of the classifier's signal.

Conclusions:

• DiagnFlow offers a novel, data set-agnostic approach for automated flow cytometry analysis.
• The tool enhances interpretability and resource efficiency in clinical settings.
• The diagnFlow web app facilitates wider implementation of automated flow cytometry analysis in hematology.