The UI was designed using HTML5 (Hyper Text Markup Language) and CSS (Cascading Style Sheets). All data for ARMOUR was stored in MySQL v5.6 database, which is a Relational Database Management System (RDBMS) and the most preferred choice for biological databases. In ARMOUR the UX is powered by Hypertext Preprocessor (PHP) and jQuery, while SQL queries were optimized for memory efficient data retrieval and specialized scripts were used for hassle free database updating. The application programming interface (API) is unique feature of the database design and schema (Figure 1). ER (Entity Relationship) tool was used to create an ER diagram that places a total of 13 tables representing comprehensive and heterogeneous information related to miRNA and mRNA that are seamlessly connected in ARMOUR. The UI or UX was designed in four different ways to interact with the database to examine the miRNA:mRNA interaction with expression level information readily accessible (Figure 2). ARMOUR database is designed to be both potable (multiple applications can be developed on the database) and scalable (additional data can be updated time to time) without changing the design.

The database includes miRNAs identified from seven varieties of rice under 38 different developmental and stressed conditions (Table 1) representing leaf, root, flag-leaf, and panicle tissues of seven varieties of rice grown under normal, salt stress and heat stress conditions. The cultivars used are traditional Pusa Basmati 1, dwarf Annapurna, high yielding Lalat, dry season Satabdi, heat-susceptible BPT 5206, salt-tolerant Pokkali and drought/heat-tolerant Nagina 22. The known or identified rice miRNAs were retrieved from miRBase Rel 21 and searched in the NGS libraries while the novel putative miRNAs were predicted using miRcat tool in sRNA Workbench from rice genes or transcripts (Stocks et al., 2012). The database includes 689 known and 1664 predicted novel miRNAs (Table 2). All miRNAs may not be necessarily present in each of the libraries, however, their large numbers indicate the abundance and variability of miRNAs.

Annotation data type was restricted to VARCHAR to ensure simplicity, fast query and retrieval while all images were generated on the fly through server side scripting. The nomenclature for known miRNA and their target transcripts, in ARMOUR, complies to the standards of miRBase and Rice Genome Annotation Project nomenclature (Kawahara et al., 2013). miRNA identifiers include miRBase ID, miRBase Accession and miRNA Family. All the predicted miRNAs are represented as “Novel-n” (where n is an integer ranging from 1 to 1664) designated as miRNA ID. The predicted novel miRNA ID will be updated periodically as they are validated and entered into miRBase. Using miRNA ID, the miRNA and the precursor sequences can be fetched out, information on the target transcripts can be obtained and the expression profiles of the miRNAs can be obtained. The gene or transcript identifiers that are covered include Entrez gene ID, MSU7 transcript ID, RAPDB gene ID, Uniprot ID, and RefSEQ mRNA ID. The gene or transcript annotation includes gene description, RNA type (coding or non-coding), chromosome number, gene start, gene end, orientation, transcript length, gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway information (Kanehisa and Goto, 2000).

To study the accurate expression profiles of miRNA, the read counts were normalized and provided as log2 of reads per million (RPM; fold expression) in the ARMOUR database. Data normalization was done to ensure minimal redundancy levels since miRNA:mRNA relationship is one is to many and vice versa. Fold Change (FC) analysis can be performed by selecting a reference sample or condition followed by comparison against one or more samples or conditions. The formula used for FC is same as used earlier (Campbell et al., 2015).

Transcripts targeted by both known and predicted miRNAs were identified using psRNATarget: A Plant Small RNA Target Analysis Server (Dai and Zhao, 2011) using the rice transcript dataset from Rice Annotation Project Version 7. Targets were qualified based on satisfying all of the following criteria (a) Expectation scores between 0 to 3 (Zhang, 2005), (b) Unpaired Energy Score (UPE) ranging from 0.0 to 25.0, (c) High scoring Segment Pair (HSP) size of 15 to 20 nucleotides, (d) Maximum number of transcripts targeted less than or equal to 200 hits, (e) Flanking length around target for accessibility analysis ranging from 17 nucleotides upstream and 13 nucleotides downstream around target size (Kertesz et al., 2007), and (f) central mismatch leading to translational inhibition centered between 9 to 11 nucleotide from the 5′ end of the miRNA (Brodersen et al., 2008). The information pertaining to transcripts targeted by the miRNA includes type of inhibition (cleavage or translation repression) and multiplicity (number of sites in the transcript likely to be targeted by the miRNA). Later the targets were also analyzed using PsRNA Target ver 2017. However, the 2017 version algorithm gave almost five times more target per miRNAs (∼30–40 mRNA per miRNA) in comparison to 2011 version due to the modified algorithm, indicating the possibility of including false positives (comparison results for 10 miRNAs is provided in Supplementary Table S2). So it was taken care to incorporate the targets that were common to both predictions.

The identified target transcripts were further annotated by identifying their GO categories, that include well planned terminologies, which describe the biological process (P), molecular function (F), and cellular components (C) of gene products, obtained from Rice Genome Annotation Project⁹. To find the associated pathways which are being targeted and affected by these miRNAs through their target transcripts, the information on associated KEGG pathway annotations to the selected target transcripts are also mentioned.

The behavior of the known and predicted miRNAs listed in the database can be inferred from their expression status in the various libraries representing response to different environmental conditions in different rice varieties. miRNA expression analysis can be performed by filtering the results based on the cut off for fold expression using conditional filter, either in one sample or condition or in multiple samples or conditions. Fold expression can be easily calculated and the results are provided as a flexible Fold Change (FC) cut off to identify up and down regulated miRNA. Fold expression and FC analysis results are provided in an interactive table with expression values and annotation (Figure 3). Within the database there is a provision of using the miRNAs (provided as examples) for retrieving and comparing their expression patterns. The results can be downloaded and used for presentation in many ways (Sharma et al., 2015; Goswami et al., 2017; Khan et al., 2018; Tripathi et al., 2018).

ARMOUR will not just act as a knowledge reservoir but also an analytic tool for the researchers to analyze and compare their data with the existing data. The users can send a request to add their own experimental data and compare with the available experimental datasets, thereby allowing sharing and comparision of data between research groups working on rice.

ARMOUR home page provides an interactive view to the user to become familiar with the database. Key aspects of UI include responsive front end (Device independent), floating frames, fluid design for tables and reader friendly color coding. Various aspects like pre-loaded sample identifiers for each query interface, client side validation of queries, simplified BLAST search, keyword filter on the query results, sorting functionality on table headers, interactive matrix table representation in advanced search, summary charts for GO and pathway queries, sequence level highlighting of miRNA structure and floating buttons ensure ease of navigation on all screens. The user can access any of these search options to retrieve all information related to the query. The results are reported in a unique correlation matrix format, which provides all the information related to the query. The matrix is populated with the number of gene hits that can be re-queried into the database.

ARMOUR has a unique query builder interface for the users that allows the database to be queried using unlimited keywords or in combination with a list of miRNA or transcript identifiers. Keywords can be searched on Gene description, GO and KEGG pathways and results are reported in a unique correlation matrix format. The matrix is populated with the number of gene hits that can be re-queried into the database. The matrix also shows how many genes are unique and being shared within and across multiple key words (Figure 3).

Using an ID of any miRNA the user can analyze all related data to the query in the database. The GO output includes information on the details of GO category to which selected targets belong and the number of the selected targets belonging to the particular GO. To find the associated pathways which are being targeted and affected by these miRNAs through their target transcripts, the information on associated KEGG pathway annotations to the selected target transcripts are also mentioned (Figure 4).