TRRUST: a reference database of human transcriptional regulatory interactions.

Han H, Shim H, Shin D, Shim JE, Ko Y, Shin J, Kim H, Cho A, Kim E, Lee T, Kim H, Kim K, Yang S, Bae D, Yun A, Kim S, Kim CY, Cho HJ, Kang B, Shin S, Lee I. · 2015

Read TRRUST: a reference database of human transcriptional regulatory interactions. by Han H, Shim H, Shin D, Shim JE, Ko Y, Shin J, Kim H, Cho A, Kim E, Lee T, Kim H, Kim K, Yang S, Bae D, Yun A, Kim S, Kim CY, Cho HJ, Kang B, Shin S, Lee I. (2015) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc4464350.txt.

Introduction

Heonjong Han and a large team of co-authors built TRRUST, a database of human gene control links, in Scientific Reports in 2015. Rebuilding the networks by which transcription factors switch target genes on or off is a long-standing goal in genetics. Computer methods that guess such links from large data sets need a trusted set of known links to be judged against, and hand-checked collections were small.

TRRUST was assembled by mining sentences in millions of medical abstracts and then curating the extracted claims by hand. It records thousands of interactions between hundreds of transcription factor genes and their targets, along with whether each factor activates or represses its target. Tools in the database test whether a factor's targets form modules, whether factors cooperate on a target, and which associations deserve priority. The authors present it as the largest public hand-curated set of its kind then available.

Catalog record

CREATORHan H, Shim H, Shin D, Shim JE, Ko Y, Shin J, Kim H, Cho A, Kim E, Lee T, Kim H, Kim K, Yang S, Bae D, Yun A, Kim S, Kim CY, Cho HJ, Kang B, Shin S, Lee I.
YEAR2015
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC4464350

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

TRRUST: a reference database of human transcriptional regulatory interactions By Heonjong Han, Hongseok Shim, Donghyun Shin, Jung Eun Shim, Yunhee Ko, Junha Shin, Hanhae Kim, Ara Cho, Eiru Kim, Tak Lee, Hyojin Kim, Kyungsoo Kim Published in: Scientific Reports DOI: 10.1038/srep11432 ABSTRACT The reconstruction of transcriptional regulatory networks (TRNs) is a long-standing challenge in human genetics. Numerous computational methods have been developed to infer regulatory interactions between human transcriptional factors (TFs) and target genes from high-throughput data, and their performance evaluation requires gold-standard interactions. Here we present a database of literature-curated human TF-target interactions, TRRUST (transcriptional regulatory relationships unravelled by sentence-based text-mining, http://www.grnpedia.org/trrust), which currently contains 8,015 interactions between 748 TF genes and 1,975 non-TF genes. A sentence-based text-mining approach was employed for efficient manual curation of regulatory interactions from approximately 20 million Medline abstracts. To the best of our knowledge, TRRUST is the largest publicly available database of literature-curated human TF-target interactions to date. TRRUST also has several useful features: i) information about the mode-of-regulation; ii) tests for target modularity of a query TF; iii) tests for TF cooperativity of a query target; iv) inferences about cooperating TFs of a query TF; and v) prioritizing associated pathways and diseases with a query TF. We observed high enrichment of TF-target pairs in TRRUST for top-scored interactions inferred from high-throughput data, which suggests that TRRUST provides a reliable benchmark for the computational reconstruction of human TRNs. Transcription factors (TFs) are major molecules that control the transcriptional activity of genes. The human genome is estimated to encode approximately 2,000 TFs1, which operate programs that change cellular states by binding to proxy or distal cis-regulatory elements (CREs) for a set of target genes. The reverse engineering of transcriptional regulatory networks (TRNs) by inferring interactions between TFs and target genes has been a key challenge in understanding the genetic regulation of complex human phenotypes. However, genome-scale regulatory circuit models remain inadequate due to the intrinsic complexities of human transcriptional regulatory programs as well as technical limitations in mapping regulatory interactions. Interactions between TFs and CREs of target genes are generally detected by DNA-binding experiments such as chromatin immunoprecipitation (ChIP), which is often followed by microarray analysis (ChIP-chip) or deep-sequencing analysis (ChIP-seq). TF-target regulatory interactions also can be inferred from high-throughput gene expression data using a wide variety of computational algorithms2. Expression data have advantages over DNA-binding data in the coverage of diverse cellular contexts, which reveal disparate sets of regulatory interactions. Thus, by integrating TF-target interactions inferred from a wide variety of cellular contexts, we may effectively reconstruct genome-scale TRNs. The integrative approach of TRN modelling requires gold-standard TF-target interactions to benchmark inferred regulatory interactions from different data sets. The reconstruction of TRNs via the integration of regulatory interactions inferred from multiple data sets has previously been demonstrated in several model organisms, and databases for literature-curated TF-target interactions have played critical roles, e.g., RegulonDB3 for an Escherichia coli TRN4, the Yeast Proteome Database (YPD)5 for a Saccharomyces cerevisiae TRN6, and the Regulatory Element Database for Drosophila (REDfly)7 for a Drosophila melanogaster TRN construction8. Several public databases for human TF-target interactions are currently available, including TFactS9, TRED10, HTRIdb11, and ORegAnno12. However, some of the databases include interactions inferred from high-throughput experiments, which may not be optimal for benchmarking. If we count only literature-curated interactions from those databases, then the size of the gold-standard data set may not be large enough to fairly evaluate TRNs. Therefore, we need to expand the size of our current set of gold-standard regulatory interactions for the more effective reconstruction of human TRNs, which is our main motivation for developing an extensive database of literature-curated human regulatory interactions. TRRUST (transcriptional regulatory relationships unravelled by sentence-based text-mining) is a database of TF-target regulatory interactions identified via the manual curation of Medline abstracts. For efficient curation over a large number of Medline abstracts, we used a sentence-based text-mining approach in which text sentences that might pertain to transcriptional regulation were first extracted and then subjected to manual curation. The current version of TRRUST contains 8,015 TF-target interactions, which to our knowledge is the largest public database of literature-curated human regulatory interactions to date. Moreover, a majority of the interactions have annotations for mode-of-regulation (i.e., activation or repression). In addition, by incorporating functional or physical protein interaction networks, TRRUST performs a network analysis of TFs and their targets to provide various systemic context information for facilitating functional interpretations, including target modularity13, TF cooperativity14, and TF-pathway and TF-disease associations. Importantly, we observed that gene pairs in TRRUST are highly enriched among the top-ranked regulatory interactions inferred from high-throughput expression data, which suggests that the TRRUST data will be a useful benchmark for the computational reconstruction of human TRNs. RESULTS TF-target interactions of the TRRUST database The overall process of constructing the TRRUST database is summarized in Fig. 1a. To increase the efficiency of the literature curation, we employed a ‘sentence-based text-mining’ approach, which is described in more detail in Methods. Briefly, we scanned ~20 million abstracts from the Medline2014 database for studies involving human biology using the MeSH descriptor ‘Humans’, which returned 7,740,270 abstracts. We then extracted 57,360 sentences that contained at least one TF name and additional gene names, which are referred to as ‘candidate sentences’. The list of TF genes were derived from Ravasi et al.15, which reported manually curated TF genes from several sources: i) the TRANSFAC database; ii) genes annotated by the Gene Ontology (GO) term ‘transcription factor’; iii) genes that contain the word ‘transcription’ in the Entrez description field; and iv) manually curated TF genes by Roach et al.16 After further curation, we generated a list of 1,984 TFs for our database. False positives of the TF list would not affect the quality of our database, because TF-target interactions will be identified by manual curation. For the given candidate sentences, we conducted a two-step text-mining procedure. In the first step, we established gold-standard candidate sentences via manual curation. This gold-standard set was updated by incoming sentences from post-manual curation, and used to prioritize incoming candidate sentences for the next round of manual curation. In the second step, we prioritized the remaining candidate sentences by a score based on the frequency difference of each word between the gold-standard positives (i.e., sentences that contain a TF and other genes for a regulatory interaction) and negatives (i.e., sentences that contain a TF and other genes but not for a regulatory interaction) (see Methods for details). We then continued the manual curation for an additional 6,000 candidate sentences from the top-scored sentences. In total, we identified 8,015 TF-target regulatory interactions between 748 TFs and 1,975 non-TF genes over two rounds of manual curation from 23,409 candidate sentences corresponding to 20,317 abstracts. The 6,000 sentences that were identified in the second round of manual curation can be used to update the set of gold-standard candidate sentences to further improve the retrieval rate in future manual curations by re-prioritizing sentences. We found that TRRUST has substantially more literature-curated (LC) human TF-target regulatory interactions than other public databases: TFactS9, TRED10, HTRIdb11, and ORegAnno12 (Table 1). TRRUST contains an approximately 2.5-fold greater number of TFs and two-fold greater number of TF-target interactions than the second largest database, TFactS.

Read the complete text