gEM/GANN: A multivariate computational strategy for auto-characterizing relationships between cellular and clinical...

Area of Science:

• Computational Biology
• Immunology
• Data Science

Background:

• Flow cytometry generates complex datasets, challenging traditional analysis methods.
• Accurate prediction of HIV disease progression is crucial for patient management.

Purpose of the Study:

• To develop and evaluate multivariate computational approaches for analyzing HIV-infected flow cytometry data.
• To improve the prediction of disease progression and survival time using machine learning techniques.

Main Methods:

• Utilized unsupervised and supervised learning on HIV flow cytometry datasets from the FlowCAP-IV Challenge.
• Applied Expectation Maximization for data preprocessing and Genetic Algorithm-Neural Network for feature extraction.
• Validated feature set reliability using WEKA-implemented classifiers.

Main Results:

• Achieved high sensitivity and specificity in discriminating HIV progressors and non-progressors using the selected feature set (TPR=1.00, FPR=0.033).
• The feature set demonstrated strong predictive capacity for survival time, particularly with unstimulated data (r=0.825).
• Statistical analysis showed promising, though variable, prediction accuracy for survival time in the test set.

Conclusions:

• The developed multivariate computational strategy effectively extracts valuable information from complex flow cytometry data.
• This approach offers a promising alternative to traditional gating for analyzing HIV progression and predicting patient outcomes.
• Further refinement may address dataset imbalances and variations for enhanced predictive performance.