Figure 4 . Overlap between Selected Features. We queried the 20 most important features in each trained classifier, SVM (Orange, bottom left), GLMNET (Purple, bottom right) and RFE-RF (Green, top), which were then compared for overlapping subsets of genes. There were 16 genes selected both by RFE-RF and SVM, with 2 genes (GB41392, GB49478) selected by all 3 approaches. GLMNET showed little overlap with SVM and RFE-RF.
Table 2 . Focal Genes. The annotations of overlapping genes (selected by at least 2 approaches) were obtained using NCBI or BLAST search (where needed). Additionally, Gene Expression Analysis (GLM and LRT) detected overlapping genes, which is indicated in the “Gene expr. Analysis” column. The “Reference” column indicates key studies that focused on the selected genes. The obtained list of focal genes included promising genes (e.g., GB49478, GB50290) that could play a key role in the dance behaviour observed in honeybees.