Genes Identified as Key Predictors
There were 18 genes (predictors) that were shared between at least two
approaches (see Figure 4). The largest overlap was observed between RFE
and SVM (16 genes) while the overlap between SVM and GLMNET was smaller
(4 genes). No overlap was detected between RFE and GLMNET. The Jaccard
Index between SVM and RFE was the most significant (0.739), while
between SVM-GLMNET and GLMNET-RFE it was 0.111 and 0.052, respectively.
Similarly, the odds ratio indicated strong association between SVM and
RFE (10265.036, p < 0.001). Overall, elements in common
corresponded mainly to protein coding genes, with the exception of
“GB40714” which corresponds to non-coding RNA. We were able to
retrieve functional information for most of the genes from annotations
of the honeybee genome (see Table 2), with the exception of “GB50940”
and” GB45448”, where annotations were available only for closely
related insects (Apis dorsata and Apis cerana ,
respectively), and “GB54617” that we could not find any information
for.