PMC:5369021 / 2257-7052 JSONTXT

{"project":"2_test","denotations":[{"id":"28347313-11040209-14906422","span":{"begin":492,"end":493},"obj":"11040209"},{"id":"28347313-22868264-14906423","span":{"begin":611,"end":612},"obj":"22868264"},{"id":"28347313-20703300-14906424","span":{"begin":752,"end":753},"obj":"20703300"},{"id":"28347313-19355820-14906425","span":{"begin":855,"end":856},"obj":"19355820"},{"id":"28347313-12610534-14906426","span":{"begin":879,"end":880},"obj":"12610534"},{"id":"28347313-22868264-14906427","span":{"begin":931,"end":932},"obj":"22868264"},{"id":"28347313-10722927-14906428","span":{"begin":1037,"end":1038},"obj":"10722927"},{"id":"28347313-16557279-14906429","span":{"begin":1040,"end":1041},"obj":"16557279"},{"id":"28347313-17540862-14906430","span":{"begin":1624,"end":1626},"obj":"17540862"},{"id":"28347313-25074712-14906431","span":{"begin":1696,"end":1698},"obj":"25074712"},{"id":"28347313-11752295-14906432","span":{"begin":1851,"end":1853},"obj":"11752295"},{"id":"28347313-18772890-14906433","span":{"begin":1888,"end":1890},"obj":"18772890"},{"id":"28347313-22955619-14906434","span":{"begin":1938,"end":1940},"obj":"22955619"},{"id":"28347313-15637633-14906435","span":{"begin":2049,"end":2051},"obj":"15637633"},{"id":"28347313-15693947-14906436","span":{"begin":2101,"end":2103},"obj":"15693947"},{"id":"28347313-20219943-14906437","span":{"begin":2105,"end":2107},"obj":"20219943"},{"id":"28347313-8594589-14906438","span":{"begin":2224,"end":2226},"obj":"8594589"},{"id":"28347313-14681366-14906439","span":{"begin":2239,"end":2241},"obj":"14681366"},{"id":"28347313-16381832-14906440","span":{"begin":2277,"end":2279},"obj":"16381832"},{"id":"28347313-19150482-14906441","span":{"begin":2423,"end":2425},"obj":"19150482"},{"id":"28347313-17299415-14906442","span":{"begin":2561,"end":2563},"obj":"17299415"},{"id":"28347313-18047712-14906443","span":{"begin":2608,"end":2610},"obj":"18047712"},{"id":"28347313-17903286-14906444","span":{"begin":2747,"end":2749},"obj":"17903286"},{"id":"28347313-23269463-14906445","span":{"begin":2890,"end":2892},"obj":"23269463"},{"id":"28347313-25433699-14906446","span":{"begin":2970,"end":2972},"obj":"25433699"},{"id":"28347313-24336805-14906447","span":{"begin":2974,"end":2976},"obj":"24336805"},{"id":"28347313-16723010-14906448","span":{"begin":4667,"end":4669},"obj":"16723010"},{"id":"28347313-16723010-14906485","span":{"begin":4667,"end":4669},"obj":"16723010"}],"text":"Background\nGene regulation is one of the most important biological processes in living cells. It is indispensable for adapting to changing environments, stimuli, and developmental stage and plays an essential role in the pathogenesis and course of diseases. Mechanistically, the transcription of DNA into RNA is predominantly controlled by a complex network of transcription factors (TFs) (see Fig. 1). These proteins bind to enhancer or promoter regions adjacent to the genes they regulate [1], which may enhance or inhibit the recruitment of RNA polymerase and thereby activate or repress gene transcription [2]. Gene products also can be modified post-translationally via microRNAs (miRNAs) degrading the transcript or inhibiting their translation [3]. Besides, a multitude of other mechanisms influence gene regulation, such as chromatin remodelling [4], epigenetic effects [5], and compound-building of transcription factors [2]. Distortion of regulatory processes is inflicted with various diseases [6, 7], especially with cancer [8, 9].\nFig. 1 Transcription of DNA into RNA. Transcription factors (TFs) bind to distal or proximal TF binding sites (TFBS) enhancing the binding of RNA polymerase and activating the transcription of DNA into RNA \nDue to this importance, many efforts have been devoted to the elucidation of human regulatory relationships and networks. Wide-spread experimental techniques are transcriptome measurements to quantify gene and transcription factor co-expression [10], chromatin immunoprecipitation (ChIP) on chips or followed by sequencing for identifying binding patterns of specific TFs [11], and bisulfite sequencing to find epigenetic signals of regulation [12]. Many large-scale datasets of such experiments have been published and are available in public repositories such as the Gene Expression Omnibus (GEO) [13], the Cancer Genome Atlas (TCGA) [14] or the Encyclopedia of DNA Elements (ENCODE) [15]. Computational methods are also used, for instance, to identify transcription factor binding sites (TFBS) [16] or to find known TFBS within the genome (e.g., [17, 18]). Several databases have been created which store relevant information, such as lists of binding motifs (TRANSFAC [19] or JASPAR [20]) or targets of regulatory miRNAs [21].\nSuch measurements and predictions are used by network reconstruction algorithms to predict regulatory relationships and regulatory networks [22]. A plethora of different methods have been proposed, ranging from purely qualitative methods [23] over simple statistical approaches [24] to more advanced probabilistic frameworks [25]. Early methods were plagued by insufficient data and a general scarcity of background knowledge, which led to rather unstable results [26]. This situation has changed dramatically over the last years, as results of more and more large screens have been made publicly available [27] and also the knowledge on principal regulatory relationships has increased [28, 29]. This, in turn, has increased the interest in methods which predict genome-wide networks using a systematic, unified, mathematical framework.\nHere, we review five rather recent methods and conduct a quantitative comparison of their results with the goal to identify their mutual strengths and weaknesses. They all have in common that they assume both the set of regulators (transcription factors or micro RNAs) to be known and the topology of the regulatory network to be given. By combining this background knowledge with specific omics data sets, especially transcriptome data, they try to infer the activity of regulators in a certain experimental condition or disease using mathematical optimization. All presented methods are global methods in the sense that they compute activities genome-wide (as much as represented by the underlying network), thus removing the shortcomings of local methods which ignore cross-talk between sub-models and global effects within samples. The methods predominantly produce a ranked list of regulators, sorted by their activity in a given group of samples; given that a multitude of biological influences is ignored during inference, especially kinetic and temporal effects, their goal cannot be to produce absolute snapshots of regulatory activity. We describe each method in detail and compare them with respect to the most important properties, such as the data being used, the method applied for deriving optimized activity values, or the evaluation performed to show effectiveness. We further implemented a quantitative comparison including four of the presented methods to objectively analyze their results. As contrast, we also include ARACNE [30] as sixth method; this algorithm uses only local reasoning and requires no background knowledge, but is still rather popular."}