Large collections of data in studies on cancer such as leukaemia provoke the necessity of applying tailored analysis algorithms to ensure supreme information extraction. In this work, a custom-fit pipeline is demonstrated for thorough investigation of the voluminous MILE gene expression data set. Three analyses are accomplished, each for gaining a deeper understanding of the processes underlying leukaemia types and subtypes. First, the main disease groups are tested for differential expression against the healthy control as in a standard case-control study. Here, the basic knowledge on molecular mechanisms is confirmed quantitatively and by literature references. Second, pairwise comparison testing is performed for juxtaposing the main leukaemia types among each other. In this case by means of the Dice coefficient similarity measure the general relations are pointed out. Moreover, lists of candidate main leukaemia group biomarkers are proposed. Finally, with this approach being successful, the third analysis provides insight into all of the studied subtypes, followed by the emergence of four leukaemia subtype biomarkers. In addition, the class enhanced DEG signature obtained on the basis of novel pipeline processing leads to significantly better classification power of multi-class data classifiers. The developed methodology consisting of batch effect adjustment, adaptive noise and feature filtration coupled with adequate statistical testing and biomarker definition proves to be an effective approach towards knowledge discovery in high-throughput molecular biology experiments.Electronic supplementary materialThe online version of this article (doi:10.1007/s12539-017-0216-9) contains supplementary material, which is available to authorized users.