Modelo para identificação de genes bimodais associados ao prognóstico no câncer

Abstract

In the last decades the biological interest in understanding gene regulation has led to the discovery of tumor genes with differentiated expression in subgroups of patients. These genes have a bimodal profile of expression value distribution, which has raised attention to investigate the patterns of development and their functionality. To better understand the bimodal pattern of these genes, the main objective of the work was to identify distinct groups of patients in a given tumor type, who had low and high levels of expression for the same gene, associated with a better or worse cancer survival prognosis. We developed a method that selects candidate genes for the bimodality pattern from the probability density function of the expression values. We analyzed 25 tumor types available in The Cancer Genome Atlas (TCGA), à we performed survival analysis using clinical information extracted from cBioPortal for Cancer Genomics. We used Fragments by Exon Kilobase per Millions of Mapped Fragments (FPKM) expression data for 24,456 genes, and found in the 25 tumor types 554 unique bimodal genes, of which 46 showed bimodal expression in more than one cancer type, with higher prevalence on the Y chromosome. The tumors KIRC, KIRP, LGG, SKCM, THCA and THYM showed consistent samples regarding survival prognosis with p-value ≤ 0.01. The method proved efficient in reducing the levels of internal variability of the groups, especially when analyzing the data by cancer subtype. As a contribution, we present a method with a free code that makes it possible to reduce the levels of internal variability of the groups and that relates the bimodal expression pattern with the survival prognosis. Thus, we believe that the use of the method may be useful in the evaluation of the bimodal pattern of gene expression and in the discovery of new clinical biomarkers for different types of cancer.