Application of iDEG to Triple Negative Breast Cancer (TNBC) Study

The method iDEG was applied to a triple negative breast cancer (TNBC) dataset queried from TCGA, which has been described earlier in Section \ref{sec:data-example}. Recall this single-subject RNA-Seq data provide measures of a breast tumor transcriptome and a surrounding healthy tissue transcriptome of a TNBC African American patient (Patient ID: TCGA-GI-A2C9). The goal of this study is to apply iDEG for the discovery of individualized DEGs for this single patient.
The expression counts are assumed to follow the NB distribution and were normalized as described in Section \ref{sec:org13dd3ad}. Since there is no prior evidence suggesting the constant dispersion across all genes, a smoothing spline (Section \ref{sec:smooth-spline}) was fit to estimate the relationship between \(\delta_{g}\) and \(\mu_{g}\). The empirical null obtained by iDEG is \(N({\color[rgb]{0,0,0}0.065,0.837^{2}})\). A local false discovery rate (fdr ) was produced for each of the 20,501 genes. Figure \ref{fig:tnbc} indicates that DEGs are determined with an adaptive cutoff that accounts for the high noise of the lowly expressed genes. For patient TCGA-GI-A2C9, iDEG identified 1,430 DEGs (approximately 7% of all genes) by controlling a local fdr below 20%. Table \ref{table:tnbc} displays the top 10 DEGs detected by iDEG, in the ascending order of a local fdr. In contrast, edgeR identified 9,921 genes as DEGs (FDR \(\leq\) 0.1), which amounts to almost half of the transcriptome. DESeq, on the other hand, only identified 194 genes (FDR \(\leq\) 0.1), which is far fewer than one would expect from a cancer patient. While it is impossible to know which genes are truly differentially expressed, the range of the number of DEGs in cancer patients is common knowledge for cancer researchers \cite{koboldt-2012-compr-molec}. We conclude that iDEG can identify a reasonable number of DEGs for this patient.
Let us now take a careful look at the genes listed in Table \ref{table:tnbc}. Adiponectin (gene product of gene ADIPOQ) and leptin (gene product of LEP) are considered mediators for the association of breast cancer with obesity, a major risk factor for breast cancer \cite{grossmann-2010-obesit-breas-cancer}. It has been shown that the reduction in adiponectin and leptin levels increases breast cancer risk \cite{miyoshi-2003-association,duggan-2011-assoc-insul,karim-2016-low-expres}, and the treatment of adiponectin induces growth arrest and apoptosis of breast cancer cell lines \cite{jarde-2009-invol-adipon,kang-2005-adipon-induc}. Our finding of the decreased expression of ADIPOQ and LEP suggests that obesity may substantially contribute to this patient’s cancer development. On the other hand, although PLA2G2A has not been extensively studied in breast cancer, many studies have shown that it inhibits invasion and metastasis of gastric and colon cancer \cite{ganesan-2008-inhib-gastr} and may predict survival \cite{xing-2011-phosp-a2}. Informed by her individualized DEG, we speculate that successful treatments for gastric and colon cancer may benefit patient TCGA-GI-A2C9. \citeauthorbubnov-2012-hypermethylation (\citeyearbubnov-2012-hypermethylation) has demonstrated the down-regulation of TUSC5 induced by DNA methylation in breast cancer. In contrast to mutated genes, DNA methylation is reversible. If the TUSC5 of patient TCGA-GI-A2C9 is suppressed by DNA methylation, pharmacologic inhibition of methylation-mediated TUSC5 suppression could potentially treat this patient \cite{baylin-2005-dna-methy}. Futher investigation of these discovered DEGs may unveil this patient’s disease etiology, progression, and possible therapeutic targets, which can eventually lead to an improved personalized treatment plan.