For the development of the AGRDB database, complete genome sequence data for A. hypochondriacus from Phytozome V13 (https://phytozome.jgi.doe.gov/Ahypochondriacus_er), A. tuberculatus, A. hybridus, and A. palmeri from a Comparative Genomics Platform CoGe: (https://genomevolution.org/coge/; Genome ID 54057, 57429, 56750), and A. cruentus was downloaded from NCBI database (https://www.ncbi.nlm.nih.gov/genome/109717?genome_assembly_id=1765032). The assembled genomes of A. hypochondriacus, A. tuberculatus, A. hybridus, and A. palmeri were scaffold-wise, and A. cruentus as 17 pseudomolecules, including chromosomes 2A and 2B ( Table 1 ). The quality of the assembled genome sequences used in this study was checked using Long Terminal Repeat (LTR) Assembly Index (LAI) method (Mokhtar et al., 2023). The genomes of A.hypochondriacus (GCA_00753965.2; LAI score: 12.81), A. tuberculatus (Genome ID: 54057; LAI score: 6.72), A. hybridus (Genome ID: 57429; LAI score: 9.56), A. palmeri (GCA_025765695.1; LAI score: 4.99), and A. cruentus (GCA_019425755.1; LAI score: 14.42), respectively.

The whole genome sequences of A. hypochondriacus, A. tuberculatus, A. hybridus, A. palmeri, and A. cruentus were used as input to the microsatellite identification tool Krait v1.3.3 (https://github.com/lmdu/krait) (Du et al., 2017) to identify SSR markers using the parameters: mono-nucleotide repeats were not considered; the minimum number of repeats allowed for the di-nucleotide were six repeats; and for tri-nucleotide to hexanucleotide, five repeats (Paliwal et al., 2016; Singh et al., 2022). The maximum difference between the two SSRs is 100 bp. Primer pairs were generated from the 250-bp flanking region of each identified SSRs using Primer3 software (Untergasser et al., 2012) with the parameters: primer length = 20–25 bp, PCR product size = 100–300 bp, with an optimum of 180 bp; annealing temperature = 65°C; GC content of (40%–60%) with optimal value 50%. The genomic coordinates (exons, introns, 3′UTRs, 5′UTRs, and intergenic region) were extracted from the gff file and genome using bed tool utilities (Quinlan and Hall, 2010).

High-quality SNP mining involves multiple steps using various bioinformatics tools. The BioProject numbers PRJNA290898, PRJNA432348, and PRJNA626536 have been used for SNP mining, and the details of SRA accessions are provided in Supplementary Table S1 . After a quality check by the FASTQC toolkit, trimming of raw reads was performed using Trimmomatic (Bolger et al., 2014). The resulting high-quality reads were mapped to the corresponding amaranth reference genomes using the BWA aligner with default parameters (Li and Durbin, 2009). The read-mapped alignment file is in SAM format. SAMTools was used to convert SAM to BAM file format, and the BAM file was then shortened by removing duplicate reads (Li et al., 2009). SNP calling and filtering were performed using SAMTools, VarScan, and bcftools. The criteria for filtering the SNPs were Q score ≥ 30, and X-coverage ≥ 30×. We used an in-house Perl script to extract the final high-quality SNPs and their location within the gene region, as well as 3′ and 5′ flanking sequences. The identification of candidate genes for transcription factor families in A. hypochondriacus, A. tuberculatus, A. hybridus, A. palmeri, and A. cruentus was performed by BLASTX similarity search of the protein-coding genes of the respective amaranth species against available peptide sequence of plant transcription factor database (PlantTFDB v4.0) (Jin et al., 2016), with the parameters bit score >100 and e-value 1 × 10⁻⁵ (Singh et al., 2017). All the identified TFs were categorized into 57 different families, and their functional annotation was performed using InterProScan 5 (Jones et al., 2014).

In silico prediction of the potential miRNAs was performed using BLASTN search of a set of mature miRNA sequences against the A. hypochondriacus, A. tuberculatus, A. hybridus, A. palmeri, and A. cruentus genome sequences. A total of 7,043 nonredundant mature miRNA sequences retrieved from the PMRD-plant microRNA database (Zhang et al., 2009) were used as the database in BLASTN search with the following parameters: (i) word match size, 7; (ii) length of mature miRNA sequence, ≥ 18 nucleotides without gap; (iii) mismatch range, 0–2; and (iv) e-value, 0.1. The sequences with less than two nucleotide mismatches were allowed, and 200 nucleotides upstream and downstream from the matched region were extracted for each miRNA (pre-miRNA sequences/possible miRNA precursors). Furthermore, sequences coding for proteins were eliminated, whereas the noncoding regions were considered for secondary structure prediction and validation. The MFold web server (http://www.unafold.org/) was used to generate and evaluate the potential secondary structures of pre-miRNAs. Screening of pre-miRNAs with the presence of proper stem-loop hairpin structure, and minimum free energy (MFE) values were calculated for each secondary structure and filtered for the secondary structure of pre-miRNAs with the lowest MFE values (Sharma et al., 2019). The pipeline for in silico miRNA prediction used in this study has been presented in Supplementary Figure S1 . Target identification of the predicted candidate miRNAs was done using the online tool psRNAtarget (http://plantgrn.noble.org/psRNATarget/) (Dai et al., 2018), using the option “Submit small RNAs and target” with default parameters. To identify transporter genes in A. hypochondriacus, A. tuberculatus, A. hybridus, A. palmeri, and A. cruentus, the protein sequences of respective amaranth species were subjected to BLASTP search against the transporter classification database (TCDB; http://www.tcdb.org/) (Saier et al., 2009), with e-value and bit score cutoffs of 10⁻⁵ and 100, respectively. The presence of the transmembrane domain was analyzed using TMHMM server v2.0 (Sonnhammer et al., 1998). The transporter genes with > 2 transmembrane (TM) domains were filtered out, and the classification of such genes in different subfamilies was done using the TCDB database.

The AGRDB database is a relational and interactive online database based on a “three-tier schema architecture” with client, server, and database tiers. The interactive, user-friendly interface of the AGRDB database has been designed using server-side programming languages such as ASP.NET, JavaScript, and Microsoft SQL Server to store large database tables. The AGRDB database contains step-by-step graphical tutorials to help users make better use of the database and a local NCBI BLAST server for nucleotide and protein sequence similarity searches. A schematic workflow of the AGRDB database showing the data generation pipeline has been presented in Figure 1 , and the database schema is represented in Supplementary Figure S2 .

The AGRDB database comprises detailed genomic information on five amaranth species, essential for the genetic improvement of the amaranth species. Data have been stored in the MSSQL Server Database in separate tables named, (i) Gene_details contains general information about the genes, such as gene ID, scaffold/chromosome number, genomic position, function, and sequence; (ii) SSR_details contains general information about SSRs, such as SSR type, SSR motif sequence, motif repeat, SSR length, and their three pairs of primer information; (iii) SNP_details stores general information about SNPs present in genic regions (exon, intron, 3′UTR, 5′UTR), as well as nongenic regions with 100 bp flanking sequences of both 3′ and 5′ regions; (iv) TF_details contains details about TFs and their annotation information, (v) miRNA_details stores general information about miRNAs, targeted genes, and their annotation information, and (vi) Transporter_details contains general information about transporter genes mentioned in Table 2 .

The AGRDB database comprises a user-friendly web interface and multiple search options for systematic data retrieval. The database has been stratified into 10 different tabs, including home, species, gene search, SSRs, SNPs, TFs, miRNAs, transporters, tools, and help and support ( Figure 2A ). The “home” page provides an overview of the database and about nutritional as well as medicinal importance of grain amaranth. The “species” tab is redirected to the separate amaranth species page that gives a short description of selected species and links to access the genomic information of that species ( Figure 2B ). The “SSRs,” “SNPs,” “TFs,” “miRNAs,” and “transporters” pages contain different search functions such as unique ID, scaffold or chromosome number, motif sequence, annotation keywords, etc. By using these search criteria, users can retrieve the information as per their requirements. Under the “tools” tab, gene expression, the local BLAST server, and JBrowse have been configured. The “help and support” tab contains a downloads page, a tutorial to access the AGRDB database more efficiently, and contact information for the users to send their queries or suggestions if they face any difficulty in accessing the database.

The AGRDB database accommodates various search menus under different sections for data retrieval and annotation information.

A brief overview of the data stored in the AGRDB database is presented in this section. The distributions of genes along with scaffolds are depicted in all five amaranth species ( Figure 6 ). The highest number of genes were identified in A. palmeri (48,625), followed by A. cruentus (43,382), A. tuberculatus (30,771), A. hypochondriacus (23,879), and A. hybridus (23,820), respectively. Scaffold 1 contains the maximum no. of genes, approximately (9.2%–10.4%) of the total genes and scaffold 16 possesses the least number of genes, which covers approximately (3.4%–4.2%) of the total genes ( Figure 6 ). Out of the total genes, A. hypochondriacus (18,432), A. tuberculatus (24,213), A. hybridus (19,602), A. palmeri (37,245), and A. cruentus (32,423) genes have been successfully annotated. Among the predicted SSRs, the maximum number of SSR markers was identified in A. hypochondriacus (205,567), followed by A. tuberculatus (165,246), A. palmeri (106,586), A. hybridus (102,492), and A. cruentus (78,445) ( Supplementary Table S2 ). From the total SSRs predicted among five amaranth species, dinucleotides were the most abundant, followed by tri-nucleotide, tetra-nucleotide, penta-nucleotide, and hexa-nucleotide repeats except in A. hybridus, A. palmeri, and A. cruentus, which had greater numbers of hexa-nucleotide than penta-nucleotide repeats ( Supplementary Table S2 ).

Among all five amaranth species, A. hypochondracus, A. tuberculatus, A.hybridus, A. palmeri, and A.cruentus, the SNPs were distributed asymmetrically across 16 scaffolds. In A. hypochondriacus, scaffold 6 contains the highest number of SNPs, 106,528 (11.11%), followed by scaffold 8 (9.71%), scaffold 1 (9.70%), scaffold 9 (8.37%), scaffold 5 (8.26%), scaffold 3 (8.17%), and scaffold 12 (1.54%) ( Figure 7 ). In A. tuberculatus, scaffold 13 contains the highest number of SNPs, 160,174 (10.19%), followed by scaffold 1 (10%), scaffold 6 (9.88%), scaffold 2 (8.01%), scaffold 3 (7.06%), scaffold 4 (6.93%), and scaffold 16 (3.52%) ( Figure 7 ). In A. hybridus, scaffold 1 contains the highest number of SNPs, 106,720 (10.76%), followed by scaffold 2 (9.31%), scaffold 4 (8.2%), scaffold 3 (7.14%), and scaffold 5 (6.71%). Similarly, in A. cruentus, scaffold 1 contains the highest number of SNPs, 142,684 (10.79%), followed by scaffold 2 (9.51%), scaffold 4 (7.9%), scaffold 9 (7.5%), and scaffold 3 (7.0%). While in A. palmeri, scaffold 4 contains the highest number of SNPs, 91,732 (12.06%), followed by scaffold 1 (10.54%), scaffold 2 (8.97%), scaffold 5 (7.73%), scaffold 6 (6.58%), scaffold 3 (6.57%), and scaffold 7, which contains the least number of SNPs, 16,358 (approximately 2.15%) of the total SNPs ( Figure 7 ). Based on the distribution pattern of TFs across scaffolds, the highest numbers of TFs were present in scaffold 1, followed by scaffold 2, scaffold 4, scaffold 3, and the lowest numbers of TFs were present in scaffold 16 in A. hypochondriacus, A. tuberculatus, and A. hybridus species ( Figure 8 ). In A. palmeri, the highest numbers of TFs were present in scaffold 2, followed by scaffold 4, scaffold 1, scaffold 5, and the lowest numbers of TFs were present in scaffold 7 ( Figure 8 ). While in A. cruentus, scaffold 2 had the most TFs, followed by scaffold 1, scaffold 3, scaffold 4, and scaffold 16, which had the least TFs ( Figure 8 ).

Based on the homology search, a total of 154 pre-miRNAs encoding 42 mature miRNAs belong to 27 miRNA families in A. hypochondriacus;126 pre-miRNAs encoding 36 mature miRNAs belonging to 26 miRNA families in A. tuberculatus; 121 pre-miRNAs encoding mature miRNAs belongs to 26 miRNA families in A. hybridus; 110 pre-miRNAs encoding 38 mature miRNAs belongs to 22 miRNA families in A. palmeri; and 119 pre-miRNAs encoding 36 mature miRNAs belonging to 25 miRNA families in A. cruentus were predicted ( Supplementary Table S3 ). In the case of transporter identification, the highest number of TM proteins was predicted in A. palmeri (6,984), followed by A. tuberculatus (6,443), A. cruentus (5,789), A. hybridus (5,106), and A. hypochondriacus (5,789) genomes. Out of the total predicted TM proteins, the non-redundant transporter genes carrying two or more transmembrane domains from A. hypochondriacus (2,760), A. tuberculatus (2,905), A. hybridus (2,642), A. palmeri (3,402), and A. cruentus (2,760) were selected in our analysis ( Figure 9 ).