Gebruik van gen-regulatorische netwerkanalyse voor de identificatie van coderende en niet-coderende mutaties uit kankergenomen

Identification of coding and non-coding cancer drivers using gene regulatory network analysis Regulation of gene transcription is an essential process, governing complex spatio-temporal expression patterns in every living cell. Gene regulation underlies processes such as the development of embryonic stem cells into various differentiated cell types, or the reprogramming of normal cells into cancer cells. Cancer is characterized by high intra- and inter-tumor heterogeneity arising from the accumulation of protein-coding, non-coding and regulatory aberrations, mostly affecting the transcriptional program. Such heterogeneity may reflect for instance a variable response to therapy even if patients have been diagnosed with the same cancer type. This highlights the necessity to study gene regulation, to model the dysregulation in cancer which may help to identify novel targets for cancer therapy. We developed integrative methods and approaches to analyze different types of next-generation sequencing data (transcriptome, genome and epigenome) to study gene regulation in normal and cancer cells. First, we developed a human and mouse version of a computational tool i-cisTarget. This method uses motif and regulatory track enrichment analysis to detect master transcription factors (TFs) for a given set of co-expressed genes or co-regulated genomic loci. Importantly, i-cisTarget allows one to reconstruct gene regulatory networks underlying biological processes of the specific (cancer) cell types, since it provides also the TF-bound target regions that are linked to genes. Second, we combined several computational approaches, including i-cisTarget, to decode melanoma regulatory landscapes. By integrating gene expression and regulatory data, we reconstructed gene regulatory networks underlying proliferative and invasive melanoma states. Particularly, we identified MITF and SOX10 as a master regulators of a proliferative network and TEAD and AP-1 as essential regulators of an invasive network. Importantly, we found a relevant link between the invasive network and drug resistance, highlighting the potential benefit of such integrative analyses for the identification of novel therapeutic targets. Third, with the development of μ-cisTarget, we introduced a new computational approach to identify gain-of-function cis-regulatory mutations that create new “edges” in the sample-specific gene regulatory networks. This approach combines whole-genome-sequencing data with transcriptomic and/or regulatory data. This tool allowed us to predict new candidate non-coding cancer driver mutations for ten cancer cell lines for which we sequenced the transcriptome, genome and epigenome. Fourth, we performed a TF ChIP-seq meta-analysis for the genome-wide prediction of TF-binding sites. We applied this peak calling threshold-free approach on 83 publicly available TF ChIP-seq data sets representing the binding of seven different TFs. This allowed us to identify a core set of bound regions per TF that are strongly preserved across different experiments (different cell types and conditions). These sets of recurrently bound regions were further used to discover important enhancer features and to train machine learning models to predict genome-wide TF binding sites. Overall, our studies show how useful computational methods can be to resolve challenging questions in regulatory genomics. Our integrated approaches may help to better understand the rules governing oncogenic transcription, which may be relevant to the cancer therapy.status: publishe