The turn-of-the-century advent of fully sequenced metazoan genomes brought with it the first genome annotations, which were largely confined to positions of confirmed and predicted genes, and typically housed in community-specific model-organism databases, e.g., ^[1][2][3][4][1,2,3,4]. Remarkably, over two decades later, the major databases (see [5]) are still mostly lacking annotation of non-coding regulatory sequences. These sequences include distal “cis-regulatory modules” (CRMs), a generic term encompassing such regulatory elements as enhancers, which mediate positive gene regulation; silencers, involved in negative regulation; and a growing number of additional elements that are not easily classified including PREs, super-enhancers, insulators, tethering elements, and others ^{[6][7][8][9][10][11]}[6,7,8,9,10,11].

2. Utility of REDfly

Prior to the development of REDfly, large-scale analyses of regulatory sequences were challenging to conduct, as the bulk of the existing regulatory data was distributed among hundreds of individual publications. Consequently, what few analyses were completed were performed on small and frequently biased sets of CRMs, such as a limited subset of early developmental pair-rule stripe enhancers in Drosophila, e.g., ^[12][13][14][16,17,18]. By curating these data and making them findable, accessible, interoperable, and usable (FAIR) ^[15][19], REDfly made it possible to bring statistical, computational, and comparative genomics methods to bear on their study. REDfly enabled the first-ever large-scale, relatively unbiased analysis of CRMs, which immediately revealed novel insights into CRM-sequence composition, differences among tissue-specific groups of CRMs, and an early indication of the presence of enhancer RNAs (eRNAs) as a prevalent CRM characteristic ^[16][20]. REDfly, by continuing to compile the data from hundreds and eventually thousands of individual experiments scattered throughout four decades of literature, subsequently proved instrumental in facilitating studies in a wide variety of research areas, including: Biology of CRMs. REDfly has been used to investigate the organization of TFBSs within CRMs ^[17][21] and how combinatorial binding influences CRM activity ^[18][22]. Soluri et al. ^[19][23] investigated how pioneer TFs control chromatin accessibility, and Blick et al. ^[20][24] examined the ability of CRMs to act in trans. REDfly data helped to illustrate how CRMs can have multiple functions ^[21][25], such as dual use as both enhancers and Polycomb response elements ^[22][26], or as both enhancers and silencers ^[23][27]. Interpretation of genomic data. REDfly has been critical for interpreting data from large-scale genomics projects including TF binding studies, e.g., ^[24][25][28,29] and studies of insulators ^[26][27][30,31]. A study challenging the understanding of which epigenetic marks characterize regulatory sequences depended on REDfly data ^[28][32]. REDfly has been used to study chromosome domains and chromatin “states”, e.g., ^{[29][30][31][32]}[33,34,35,36], to explore 3D-chromatin conformation ^[33][34][37,38], to study ncRNA and eRNA expression ^[35][36][39,40], and to validate scATAC-seq approaches, e.g., ^[37][38][41,42]. Computational CRM discovery. REDfly has played a dramatic role in methods for computational CRM discovery, both as a source of training data and as a method for validating predictions, e.g., ^{[39][40][41][42][43][44][45][46][47]}[43,44,45,46,47,48,49,50,51]. Su et al. ^[48][52] used REDfly data to assess CRM-discovery approaches, which would have been impossible without REDfly. Computational CRM-discovery methods using REDfly for training data also can identify CRMs in diverse insect species ^[49][53] and, as such, provide a powerful tool for annotating insect regulatory genomes ^[50][54]. CRM evolution. REDfly has enabled studies of CRM evolution and TFBS turnover, e.g., ^{[51][52][53][54][55][56][57]}[55,56,57,58,59,60,61]. Wang et al. ^[58][62] used REDfly data to investigate the selective pressure on DNA shape at TF binding sites, and Peng et al. ^[59][63] explored the relationship between chromatin accessibility and TF binding to predict evolutionary changes in enhancer activity. As can be seen from these examples, REDfly is an important source of raw data for analysis, hypothesis generation, assessment, validation, and empirical research.

3. REDfly Data Model

REDfly curates two types of data: CRMs and transcription factor binding sites (TFBSs), with CRMs being the main focus. Historically, CRM annotations have been drawn from reporter gene assays in transgenic animals or cultured cells, but an increasingly diverse set of assays are starting to be included. In particular, several years ago REDfly started to capture CRMs identified through various “X-seq” assays, such as ATAC-seq, FAIRE-seq, DNase-seq, ChIP-seq, etc., as well as from purely computational predictions, in recognition of the fact that many regulatory sequences are presently being defined by these methods. There is considerable debate in the regulatory genomics field as to just how CRMs should be defined, with several studies indicating that the different methods of CRM identification have led to widely non-overlapping results and raising questions as to which, if any, of these methods most accurately identify CRMs ^[60][61][62][64,65,66]. As a result, REDfly separates its CRM data into four distinct subclasses: reporter constructs (RC), CRM_segments, predicted CRMs (pCRM), and inferred CRMs (iCRM). RCs and CRM_segments are drawn from activity-based assays (Figure 1, left), primarily gene-expression data from either reporter genes or from native genes following mutation or deletion of regulatory sequences. RCs mainly represent reporter-gene results, assayed either in transgenic animals or in cell-culture assays; the two types of assays can be independently searched. The CRM_segment class contains sequences that are demonstrated to be necessary for gene regulation but, unlike in a reporter gene assay, are not necessarily sufficient. Such sequences can be obtained from the analysis of small chromosomal deletions or site-directed mutagenesis but are increasingly being found in the literature as a result of CRISPR/Cas9-mediated targeted sequence deletions. pCRMs, on the other hand, reflect CRMs identified by assays that do not require demonstration of activity, for example, the presence of histone modifications, or computational predictions (Figure 1, right). iCRMs are not curated from the literature, but represent putative regulatory elements based on the analysis of other REDfly data (see below).