Reactivation of silent genes in cancer

Question

[zhiyuan]

Hello everyone, I'm a graduate majoring in biology and new to bioinformatics. By browsing gene expression data in TCGA, I noticed that in cancer, some genes silent in corresponding normal tissue (very low or close-to-zero expression) were reactivated and expressed at different degrees. Intuitively, I think that it's because of the genome instability in cancer cells, and most of these reactivation do not contribute to tumorgenesis. But if there are genes reactivated frequently (~20% for example) in tumor samples, some of them may be helpful for prognosis and diagnosis. So I want to know is there any research about this issue? I tried searching on PubMed but I didn't find anyone (maybe the keywords I used were wrong). If not, how can I find an algorithm to find such genes? I know nothing about mathematical modeling and have no idea how to find them with statistics significance. Any disscussion on this topic is welcome. Thank you!

[LChart]

The re-activation or overexpression of oncogenes in cancers is well-established, and can be driven by copy number alterations or translocations (gene amplification), chromosomal rearrangements, and epigenetic de-repression.

More generally, the concept of "Epithelial-to-Mesenchymal transition" in carcinomas (and equivalent stemness programs in other cancers) precisely involves the repression of cell fate genes, and de-repression of pluripotency and renewal programs, leading to cells that can proliferate more extensively and rapidly than their progenitors.

Answer 1

[ATpoint]

Think about it, this is not a novel concept from a technical standpoint. "Reactivation" of a gene would mean that the cancer group has more counts for that gene than the control group, with the additional filter that counts in the control should be quite low or close to zero. Eventually, that is nothing else than a differentially-expressed (overexpressed) gene in the cancer group. Classifiers and prognosis, be it on differential genes, a priori-defined genes or groups of genes (for example co-expressed modules) are not new, and TCGA data have extensively been used for this. I therefore think that you have no novelty here, but I appreciate that you share your ideas as a good discussion is the basis for formulating novel project ideas.

In order to give a technical answer with some buzzwords: You would get the count matrix for your samples, scan for differential genes with frameworks such as edgeR, DESeq2 or limma-voom, and then filter for significant genes that have low counts in the control. This can essentially be done in a few lines of code. If you need feedback then please try and post where you get stuck it after reading the excellent vignettes of the tools.

[zhiyuan]

Thans for your answer, I will try those frameworks!