A hybrid ensemble feature selection design for candidate biomarkers discovery from transcriptome profiles

Discovering disease biomarkers from gene expression data has been greatly advanced by feature selection (FS) methods, especially using ensemble FS (EFS) strategies with perturbation at the ( i.e., homogeneous EFS) or method level heterogeneous EFS). Here, we proposed a hybrid EFS design that explores both types of to disrupt associations good performance single dataset, algorithm, specific combination both, which is particularly interesting for better reproducibility genomic biomarkers. We investigated adequacy our approach microarray in four cancer, extensively comparing it other and approaches. Five algorithms were analyzed: Wx, Symmetrical Uncertainty, Gain Ratio, Characteristic Direction, ReliefF. observed that, across distinct datasets, approaches attenuated large variation most methods without function perturbation. Additionally, superior EFS. Interestingly, ranks produced reached greater biological plausibility, notably high enrichment cancer-related genes pathways. Thus, experiments suggest potential discovering candidate data. Finally, provide an open-source framework support similar analyses this domains, being available as user-friendly application programmable Python package.