Computationally efficient multi-sample flow cytometry data analysis using Gaussian mixture models.

Area of Science:

• Computational Biology
• Biostatistics
• Data Science

Background:

• Comparing cell populations across multiple flow cytometry (FCM) samples is a significant challenge.
• Existing multi-sample mixture models, often Bayesian, are computationally complex and lack scalability for large FCM datasets.

Purpose of the Study:

• To develop a computationally efficient multi-sample Gaussian mixture model (MSGMM) for scalable FCM data analysis.
• To facilitate direct comparison of cell populations across heterogeneous FCM samples.

Main Methods:

• Extension of basic Gaussian mixture models (GMMs) to handle multiple samples.
• Application of an efficient expectation-maximization algorithm for model fitting.
• Development of heuristics for analyzing MSGMM output to reveal sample patterns.

Main Results:

• MSGMMs are competitive with existing models in rare cell detection and sample classification accuracy.
• The model demonstrates scalability for large FCM datasets.
• Heuristics applied to MSGMM output effectively reveal structural patterns within sample collections.

Conclusions:

• The efficient MSGMM recovers the utility of more complex models while offering scalability.
• Fitting GMMs to large FCM datasets opens avenues for discovering associations between sample composition and clinical outcomes or treatment responses.