BioProject

The repositories for various “omics” data collected from different cancer types are constantly growing. More...

The repositories for various “omics” data collected from different cancer types are constantly growing. However, robust diagnostic and/or prognostic conclusions can often not be extracted from mixed transcriptomes or other heterogenous datasets obtained from large cohorts as important signals or single features can be masked. Here, computational microdissection of bulk transcriptome data was applied to gain insights into patient prognosis and to investigate important processes and cell subtypes within new samples in silico. We developed parallel consensus ICA that decomposes many whole transcriptomes into independent signals (components), some of which originated from distinct cell subtypes while others accounted for technical biases. We show that the weight of components allows for prediction of clinically relevant patient characteristics, which was further validated on an independent patient cohort. Moreover, ICA components can be linked to biological functions, thus new samples could be classified by presence of respective biological properties such as immune signals, angiogenic activity and proliferation. Finally, through integration of different data types (transcriptomes and miRNomes) by ICA, biological functions of miRNAs were deduced, which would otherwise not be possible. Taken together, ICA represents a versatile tool to dissect complex data cohorts into individual components allowing for better use of such datasets. Overall design: 1 cell line (NHEM) and 4 biopsies have been submitted for RNA-Seq. Biopsies correspond to 3 different patients caraterized as primary melanoma tumor. Additionally, one matched normal skin has been used as control. Less...