Application of ensemble learning to the differential gene expression in left-right breast tumors

Differential gene expression analysis is becoming an increasingly popular means of understanding the genetic basis of diseases leading to more informed drug design, improved diagnostics, and personalized medicine. However, the results of various differential gene expression packages vary significantly when applied to the same dataset. Diverging results leads to an ambiguity in the identification of the culprit gene or genes, which reduces the impact of differential gene expression studies. To resolve these ambiguities, an ensemble learning technique was deployed to combine the results of several approaches in order to achieve a more reliable set of differentially expressed genes. The method was then evaluated on a triple negative breast cancer dataset to uncover the differences between left and right breast tumors that lead to a diminished prognosis of right breast cancer versus left breast cancer. The ensemble analysis incorporated the most popular differential gene expression analyses, namely DESeq2, edgeR, baySeq and DSS. While each method readily produced a list of top 200 prominent genes, the pair-wise comparison of these lists produced Pearson coefficients ranging from 0.0009 to 0.639. An ensemble-analysis approach, on the other hand, resulted in a total of eight significant genes: IYD, CRISP3, MAGEB2, DHRS2, UPK1A, PRR4, PAEP and SSUH2. Of interest are CRISP3 and MAGEB2 due to their previously documented connection to various forms of cancer.