CVID

CVID applies automated machine learning to flow cytometric immunophenotyping data to classify Common Variable Immunodeficiency (CVID) patients from other primary antibody deficiencies (PADs) and healthy controls.

Key Features:

• Automated machine learning pipeline: Performs end-to-end automation from quality control and data pre-processing to population identification (gating), feature extraction, and classifier training.
• Flow cytometric immunophenotyping input: Uses flow cytometry immunophenotypic profiles as the primary input for analysis.
• Automated population identification (gating): Replaces manual gating to reduce observer bias and detect populations beyond manually defined gates.
• Feature extraction: Extracts quantitative features from identified cell populations for downstream classification.
• Classifier performance: Achieves an average balanced accuracy of 0.93 (±0.07) compared with 0.72 (±0.23) from traditional manual gating.
• Cohort diversity: Applied to pediatric and adult CVID, idiopathic primary hypogammaglobulinemia, IgG subclass deficiency, isolated IgA deficiency, isolated IgM deficiency, and healthy controls.

Scientific Applications:

• Differential diagnosis of CVID: Distinguishes CVID phenotypes from other PADs and healthy controls using immunophenotypic signatures.
• Improved diagnostic reproducibility: Increases reproducibility and accuracy of flow cytometry–based immunophenotyping relative to manual gating.
• Phenotypic profiling of PADs: Enables identification and comparison of immunophenotypic profiles across CVID and related antibody deficiency disorders.

Methodology:

Computational steps explicitly include quality control, data pre-processing, automated population identification (gating), feature extraction from flow cytometry data, training a machine learning classifier, and evaluation using balanced accuracy metrics.