T-cell acute lymphoblastic leukemia (T-ALL) is an aggressive hematologic malignancy 
characterized by extensive genetic heterogeneity. Despite major advances in medicine towards 
personalized therapies, a significant subgroup of patients still experience relapse or treatment 
resistant disease, which is associated with poor prognosis. The aim of this project is to identify 
molecular signatures of relapse and/or treatment refractoriness across multiple omic layers and 
to develop a multi-omic predictive model for these events.

The study cohort comprised 62 pediatric patients diagnosed with T-ALL, for whom whole-genome 
sequencing (WGS), RNA-seq, miRNA-seq and DNA methylation profiling were available. Patients 
were clinically annotated by experts according to three outcome categories: relapse occurrence, 
treatment refractoriness, and the presence of either event. Each omic dataset underwent 
standard preprocessing, including quality control and removal of invariant features. For WGS and 
RNA-seq data, multiple aggregation strategies were applied to derive biologically meaningful 
feature representations. Feature selection was performed using the Boruta algorithm, and 
classification models were constructed with a Random Forest approach to identify predictive 
molecular patterns associated with relapse and refractoriness. Before classification, 
downsampling was applied to balance class sizes, and the entire procedure was executed within 
a LOOCV loop.