Clustering methods and Bayesian inference for the analysis of the evolution of immune disorders

TL;DR

This paper evaluates unsupervised clustering algorithms for diagnosing immune disorders, specifically systemic lupus erythematosus, using Bayesian analysis to determine optimal hyperparameters and assess uncertainty in method selection.

Contribution

It introduces a Bayesian framework based on the Plackett-Luce model to compare clustering strategies and quantify uncertainty in hyperparameter choices for immune disorder analysis.

Findings

Clustering algorithms can effectively detect disease flares and remission.

Bayesian analysis helps identify the most suitable clustering method for specific data.

Uncertainty quantification improves confidence in diagnosis strategies.

Abstract

Choosing appropriate hyperparameters for unsupervised clustering algorithms in an optimal way depending on the problem under study is a long standing challenge, which we tackle while adapting clustering algorithms for immune disorder diagnoses. We compare the potential ability of unsupervised clustering algorithms to detect disease flares and remission periods through analysis of laboratory data from systemic lupus erythematosus patients records with different hyperparameter choices. To determine which clustering strategy is the best one we resort to a Bayesian analysis based on the Plackett-Luce model applied to rankings. This analysis quantifies the uncertainty in the choice of clustering methods for a given problem