A recent MOLAB study aimed to validate previous results from the group that suggested the possibility of predicting recurrent leukemia in children from flow cytometry samples at diagnosis. Among the variety of approaches for processing and extracting features from this data type, this study focused on the classical median-based parameterization of surface protein expression as well as more complex descriptors of the expression distribution. A comprehensive pipeline for normalization, feature selection, clustering of cell populations, and predictive classification was built and tested on an expanded cohort of 188 patients. 

The analysis revealed minimal differences between relapse and non-relapse patients, despite evaluating a broad array of metrics and varying resolutions for cell subpopulations discovery. Comparison with existing biomarker discovery algorithms further confirmed these negative results. The study suggests that meaningful immunophenotypic differences related to relapse risk may only emerge at later stages of treatment or that alternative, potentially more complex methods of data analysis are needed to identify relapse-specific biomarkers.