Schematic of the benchmark workflow. Three multi-omics datasets, covering distinct medical applications, were selected. A reference simulation scenario was designed using SNR and sparsity levels estimated from real-world datasets. A total of 14 alternatives were also generated by modifying class imbalance, SNR, dimensionality, relative importance of effects, etc. A selection of six integrative approaches, representative of existing methods, were evaluated on both real-world and simulated data based using MCC.