Classifying flow cytometry data using Bayesian analysis helps to distinguish ALS patients from healthy controls.

Area of Science:

• Immunology
• Computational Biology
• Biostatistics

Background:

• Multidimensional flow cytometry (mFC) is crucial in immunology for analyzing cell populations.
• Traditional mFC data analysis relies on manual gating, which can be subjective and time-consuming.
• Computational methods are emerging to enhance mFC data analysis.

Purpose of the Study:

• To develop and validate a Bayesian network analysis model for classifying amyotrophic lateral sclerosis (ALS) using raw, ungated mFC data.
• To compare the performance of Bayesian network analysis against a commercial algorithm (Citrus).

Main Methods:

• A Bayesian network model was constructed using raw mFC data from healthy controls (HC) to create a reference 'HC tree'.
• This model was used to predict disease status (ALS or HC) based on marker distribution.
• The algorithm calculated the probability of zero marker distribution to assess similarity between samples and the HC tree.

Main Results:

• The Bayesian network model correctly identified 64/68 ALS cases in the primary cohort and 100% in a validation cohort.
• Optimal performance was achieved using 7 markers, 200 bins, and 20 patients (p < 0.0001).
• Bayesian network analysis demonstrated superior performance compared to the commercial algorithm 'Citrus'.

Conclusions:

• Bayesian network analysis offers a novel, data-preserving method for classifying mFC data without reduction techniques.
• This approach shows potential as a complementary diagnostic tool in clinical settings for diseases like ALS.