Hien Nguyen, M.S.
Christopher Frydenlund,Â B.S.
Mingzhou Song, Ph.D.
New Mexico State University, Las Cruces
We developed a novel computational framework to study co-regulation of cancer driver genes by transcription factors (TF) and long non-coding RNAs (lncRNA). The analysis is non-parametricâ€“assuming no mathematical form of underlying interactions. Unlike other methods based on binding sites that only determine common direct gene targets of TFs and lncRNAs, our method based on gene expression profiles can detect both monotonic and non-monotonic dependency.
We obtained transcriptome data from FANTOM5, lncRNA annotation from GENCODE, and TF from Human Transcriptional Regulation Interactions database.
For a cancer driver gene, we test its functional dependency on every possible pair of lncRNA and known TF of this target gene. If the dependency is significant, we determine the effects of ncRNA, TF or their interaction. The statistics are computed based on functional chi-square, a method to detect the functional dependency of a response variable on an explanatory variable.
We detected regulation of lncRNAs to cancer driver genes, and their interactions with TFs of the target genes. For example (Fig. 1), lncRNA MALAT1 has a positive dependency with BRCA2â€“a breast cancer gene (p=0.014), but only when the transcription factor YBX1 is highly expressed. When YBX1 is under-expressed, the dependency of BRCA2 on MALAT1 becomes insignificant (p=0.1574).
Given a lncRNA, our framework identifies its new transcriptional targets, and combinatorial interactions with transcription factors. Together with sequence and chromosome structure information, we can hypothesize on in-cis or in-trans regulation of potential lncRNAs.