Agricultural Sciences in the Big Data Era: Genotype and Phenotype Data Standardization, Utilization and Integration

1. Introduction

Genotype-phenotype (G2P) integration is the process of linking genetic data to measurable qualitative and quantitative phenotypes and traits. Historically, linking genetic markers or genes associated with desirable traits have led to the development of improved cultivars with higher yields and quality, enhanced disease resistance or climate resilience. In past two decades, the generation of high-throughput Omics or “big data'' including plant genomes and pan-genomes, genetic variation data including Single Nucleotide Polymorphisms (SNPs) and structure variations (SVs), transcriptomes, phenotype, proteomes, and metabolomes has changed the scale and scope of data analysis, knowledge synthesis and its application in translational research (1,2). In addition, researchers and breeders worldwide have collected classic mutant phenotype and trait data, and more recently the large-scale phenotype data collection is peaking by. Often, a particular set of big data is being analyzed to bridge specific knowledge gaps identified by the projects and remain underexploited for synthesis of new knowledge. Going forward, the different data types produced in various experiments can be reutilized for synthesizing new knowledge, developing data-driven hypotheses, and for experimental research. The integration of large-scale datasets from diverse sources, however, can be challenging and typically involves quality check, data re-formatting, curation, and re-analysis. For example, genotype, phenotype and expression data for the same plant accessions were generated from various projects over a decade, each using inconsistent sample identifiers and different plant growth environments. Before utilizing these various datasets to investigate the genetic and environmental factors influencing a particular phenotype, establishing consistent sample names, gene Ids, and phenotypes across all datasets will be needed and possibly require modification in the original data format. The fulfillment of the unprecedented potential of big data depends on the data being Findable, Accessible, Interoperable, and Reusable (FAIR) (3-5). To meet the FAIR standards, any dataset should include metadata providing the standard terms and details necessary for data interpretation using plant ontologies or controlled vocabularies. Data and metadata standardization can be achieved by developing common community standards of data formats, and description so that diverse datasets from different sources can be accessible and interoperable for visualization and knowledge synthesis. A clear, organized and consistent method of capturing and exchanging agricultural data will ensure easier data discovery, comparisons, and reuse by various stakeholders.

Making data FAIR requires concerted efforts and communications among all parties involved in data generation and curation. In 2015, the AgBioData Consortium (https://www.agbiodata.org) was formed to identify and promote the means to consolidate and standardize common Genetic Genomic Breeding (GGB) database tools and operations, with the goal towards increased data interoperability for future research (4). At present, AgBioData comprises over 40 GGB databases and more than 200 scientists, fostering collaborations and open discussions about the common practices, challenges and solutions to big data agricultural research. A AgBioData consortium white paper (4) has previously identified challenges facing GGB Databases and suggested common guidelines for bio-curation, ontologies, metadata, GGB database platforms, programmatic (machine) access to data, communication among various partners and stakeholders, and sustainability of genomic resources/databases. AgBioData aims to (i) identify and address data-related issues by defining community-based standards; (ii) expand the network by involving all the stakeholders of the agricultural research community; (iii) develop educational material to train current and future scientists on database usage and the FAIR principles; and (iv) develop a roadmap for a sustainable GGB database ecosystem. As part of this NSF RCN project, working groups were formed around major data-related challenges and needs. The Genotype-Phenotype working group (G2P-WG) was formed in November 2021 with the goal of identifying current challenges in annotating and integrating large-scale genotype and phenotype data. The efforts and work of the AgBioData GP-WG brought to the current white paper, which summarizes the current status of FAIR practices of phenotype and genotype data (see Figure 1). For common genotype and phenotype data, we report (i) a brief introduction of the diverse data type and how they are generated, (ii) primary and secondary data repositories and databases for these data types, (iii) requirements of associated metadata and the minimum standards, (iv) examples of re-use and reanalysis of omics data, and (v) limitations of data re-use.

2. Genomics and Transcriptomics data

2.1. Whole genome and transcriptome sequences

In the past decade, sequencing technology has evolved rapidly from the early days of time-consuming Sanger sequencing to high-throughput massive parallel sequencing that started the era of the Whole Genome Sequencing (WGS) and transcriptome sequencing. There are basically three general methods of DNA/cDNA sequencing : (i) Sanger chain termination sequencing and Maxam Gilbert sequencing; (ii) short-read sequencing known as Next Generation Sequencing (NGS) (6) including Ion Torrent, Solexa/Illumina, Roche/454 pyrosequencing; and (iii) more recent long-read Third Generation Sequencing (3GS), primarily single molecule real time sequencing from Pacific Bioscience (PacBio), and nanopore sequencing from Oxford Nanopore Technologies (ONT). At present, Illumina is the dominant and most popular platform in NGS for both genomes and transcriptome sequencing because of high accuracy, low cost, and global distribution. PacBio and ONT are gaining popularity and becoming affordable for high-quality long-read/ full-length sequences. Similarly, DNBSeq from MGI Tech, a subsidiary of Beijing Genomics Institute (BGI) group and Ion Torrent Systems (7,8) are making advances. There are several file formats used in WGS and the most common is the compressed FASTQ format that is used for both sequencing platforms, NGS and 3GS. The original file formats for 3GS include legacy h5 format for PacBio, the industry-standard BAM format, and the FAST5 format for ONT that is based on the hierarchical data format (HDF5) used for ONT data storage sequencer. There are numerous basecallers available for conversion to FASTQ format (9) and in general we find that sequencing data has achieved standard data formatting.

2.2. Genome Sequencing Strategies for Genotyping

Genotyping is a crucial component in linking genotype to phenotype. The first-generation genotyping marker was Restriction Fragment Length Polymorphisms (RFLPs), which relied upon underlying differences in base pair sequences to create an autoradiographic fingerprint after DNA regions were digested with known restriction enzymes (10). These techniques progressed with technological advancements in PCR and other DNA sequencing techniques to include genotyping via microsatellite markers, specifically simple sequence repeats (SSR) or short tandem repeats (11). High throughput low- and high- density SNP arrays provide a cost-effective genotyping solution for studies such as population structures, genomic diversity, gene discovery and molecular breeding. Array technology can genotype a large number of samples in a short period of time, and data analysis is much simpler. However, designing an efficient array with high quality SNPs for a particular crop usually requires significant investment upfront. As genome sequencing has advanced even further, researchers can now achieve whole-genome profiling through lower- or higher-coverage sequencing strategies such as NGS and 3GS.

Various genome sequencing strategies can be employed based on research aims and funding.

Sequencing of sub-sampled loci (12) has been widely used in phylogenomics studies for cost-effective large-scale genotyping. Skim sequencing (13) is a low coverage whole genome sequencing approach. Target enrichment sequencing investigates specific genomic elements via pre-defined probe sequences (14). Exome sequencing is a common type of target sequencing that focuses on protein-coding regions of genes (15). Amplicon sequencing is a highly targeted approach addressing specific genome loci. Genotyping-by-sequencing (GBS) (16,17) and restriction-site associated DNA marker sequencing (RAD-seq) (18,19) are two popular cost-effective sequencing strategies for shearing the genome via restriction enzyme(s). This advent of high-throughput sequencing has generated immense amounts of data that plays into several areas of genomic concern. Regarding genotyping data structure, the 1000 Genomes project (https://www.internationalgenome.org/) spearheaded the first Variant Call Format (VCF) for standardizing the SNPs, indels, and structural variation between two or more genomes at a given locus (20). The VCF has become the go-to format for variant data and associated metadata; over time, modifications of the base VCF file have expanded to include experiment-specific modifications, such as GWAS-VCF (21) and GVCF (tinyurl.com/5f8wpmhr), and accommodates variant information of polyploid genomes. In addition to low coverage genome sequence, transcriptome sequence is routinely used for genotyping and identification of useful genetic markers. More recently, integration of single cell genome sequencing and single cell transcriptome sequencing tools have facilitated quantifying genetic and expression variability between individual cells (22). Like sequence data, genotyping data has standardized formats.

2.3. Public repositories for genomics and transcriptomics data

Regardless of the sequencing platform or strategy used, raw sequencing data in compressed fastq.gz format is submitted to a public data repository such as National Center for Biotechnology Information (NCBI) GenBank, Sequence Read Archive (SRA) and/or Gene Expression Omnibus (GEO) via the NCBI submission portal. NCBI provides BioSample metadata templates based on organism lineage validation. Besides NCBI, the data can be submitted to the DNA DataBank of Japan (DDBJ), Sequence Read Archive (SRA) via DDBJ submission navigation website, or the European Nucleotide Archive (ENA) through BioStudies portal. DDBJ, ENA and NCBI GenBank (see Table 1 and Supplementary Table S1) form the International Nucleotide Sequence Database Collaboration (INSDC) and exchange data daily. Prior to publishing the results, all the life science journals require authors to submit their raw sequence data to the public INSDC repositories - a key component of the data sharing policies in the community of biologists (23). Additional public platforms that host the sequence data includes the US Department of Energy (DOE) Joint Genome Institute (JGI) that makes sequencing data generated by its collaborating projects available immediately to registered users and then follows public release on JGI and NCBI/ SRA or GeneBank after a one-year embargo period. JGI also provides Phytozome (24), the Plant Comparative Genomics portal, for genome accessing, comparison and visualization (see Table 2). Nature and Scientific Data request that sample metadata is deposited in one of the INSDC BioSample databases in conjunction with sequence data. It is crucial to use the standardized metadata both at the study and sample level to facilitate the curation and processing of transcriptomics data in a FAIR-compliant way. A few sequence repositories such as Zenodo (https://www.zenodo.org), DRYAD (https://datadryad.org), Figshare (https://figshare.com), Harvard Dataverse (https://dataverse.harvard.edu), etc. accepts data submission in any file format.

Outside the public databases hosted in the USA and Europe, the Genome Sequence Archive (GSA, https://ngdc.cncb.ac.cn/gsa) in China follows INSDC-compliant data standards (25). The Indian Biological Data Center (IBDC, https://ibdc.rcb.res.in) is a public repository in India to host various life science data. For sequencing data, IBDC provides the INSDC-compatible Indian Nucleotide Data Archive (INDA, https://inda.rcb.ac.in/home) with data synchronized to NCBI/ENA/DDBJ; and the Indian Nucleotide Data Archive-Controlled Access (INDA-CA, https://inda.rcb.ac.in/indasecure/home) for private data. In New Zealand, the Aotearoa Genomic Data Repository (AGDR) hosts genomics data, especially for native taonga (‘treasure’ in Maori language) species.

Data submission to cloud storage is also gaining popularity. Amazon Web Services (AWS) offers Open Data (https://aws.amazon.com/opendata) for unregistered users to find and use publicly available datasets, and allows subscribed customers to search and access third-party data (https://docs.aws.amazon.com/data-exchange/index.html). In addition, it provides Amazon Omics (https://aws.amazon.com/omics/) and Plant & Animal Genomics (https://aws.amazon.com/solutions/agriculture/plant-animal-genomics/) platforms to facilitate omics data analysis and integration. Other options include Google Cloud Life Sciences (https://cloud.google.com/life-sciences) and Microsoft Genomics (https://azure.microsoft.com/en-in/products/genomics/).

We compiled a list of public repositories for genome, genotyping and transcriptome sequence data (see Table 1) that are active, maintained and updated. Detailed information about metadata availability, data file formats related to these repositories are described in Supplementary Table S1.

The metadata associated with sequence and genotype data promotes a dataset’s discoverability and reusability. We note here that many specific secondary public repositories exist that exclusively host data on promoters, transcription factors, proteomes, various RNA types, epigenomic data and Pangenomes (see Supplementary Table S2). However, here we limit our discussion to primary genotype and phenotype data and expect that detailed discussions on other related topics will be provided by the other working groups of the AgBiodata consortium.

2.3.1. The metadata requirements on genomics and transcriptomics data set

The metadata associated with the genome/genotyping/transcriptome sequencing is crucial for its re-use and interoperability. To maximize the implementation of FAIRstandards, the metadata should be described with accurate Gene Ontology (GO) and Plant Ontology (PO) terms with proper evidence codes wherever applicable. Project and sample level metadata typically include taxonomic identifier (for species), tissue type (organism part) from which the sample was taken, disease state, growth or developmental stage of the sample, the biological gender of the sample, and collection date. Assay level metadata are directly related to the preparation of biological materials undergoing the assay including method detail(bulk RNA-seq, scRNA-seq, etc), library information (single-end or paired end), replicates (biological or technical), instrument metadata, quality control (QC) and workflow metadata. For example, submission of sequencing data to NCBI GenBank and SRA requires metadata for the submitter (i.e., name, affiliation and email of the data submitter, and other authors), BioProject goals (i.e., genome sequencing and assembly; raw sequence reads, epigenomics, exome, proteome, variation, etc.), and BioSamples information (i.e., organisms name and taxonomic identifier, geographical origin of the sample and tissue type).

We note here that in most repositories, the organism’s name is the only required field for biological targets, with optional fields of strain, breed, cultivar, isolate name, label, and description. The data release date, project title, and public description of the study goals are the minimum general information required for a project. Optional fields include a project’s relevance to a field(agricultural, medical, industrial, environmental, evolution, model organism, and other), external links to other websites associated with the study, grant information (number, title, grantee), research consortium name and URL, data provider and URL (if different from the submitting organization), publication information.

Optional but useful metadata for BioSamples include sample title, BioProject accession, biomaterial provider (lab name and address, or a cultural collection identifier), name of the cell line, cell type, collected by and date, culture identifier and source institute (refer to http://www.insdc.org/controlled-vocabulary-culturecollection-qualifier), disease name and stage, observed genotype, growth protocol, height or length measured, the growth environmental , the geographical coordinates of the sample collection, phenotype of sampled organism (compliant with the Phenotypic quality Ontology (PATO) terms at http://bioportal.bioontology.org/visualize/44601), population (filial generation, number of progeny, genetic structure), sample type (cell culture, mixed culture, tissue sample, whole organism, single cell, and so on), sex, specimen voucher, temperature of the sample at time of sampling, treatment and sample description.

The mandatory attributes for library construction metadata are BioSample name, library ID, a title, data type and method information (eg. WGA, WGS, RNA-Seq, EST, ChIP-Seq, and so on), source (GENOMIC, TRANSCRIPTOMIC, GENOMIC SINGLE CELL, METAGENOMIC, etc.), selection (PCR, RANDOM, RT-PCR, cDNA, DNAse, Restriction Digest, etc.), layout, platform, instrument model, design description, file type and filename(s).

A few other sequence data repositories do not enforce submission of metadata but encourage data submitters to provide as much details as possible. In this category AGDR (https://repo.data.nesi.org.nz/DD) requires submitter ID, project ID, project code, project name, programme name, database gap accession number, experiment type, number of samples and replicates, and data type. In addition, it provides metadata templates for submitting detailed information on samples and methods (Sample, Aliquot, RIN, adapter name and sequence, barcoding, base caller name and version, experiment name, flowcell barcode, fragment sizes, instrument model, lane number, library name, library preparation kits), project, publication, core metadata collection, indigenous governance, and indigenous knowledge label templates.

The minimum metadata for a DRYAD submission requires a title describing the data and the study, author(s) information, abstract (dataset structure and concepts, reuse potential, any legal or ethical considerations, etc.), and research domain. Optional metadata recommended are funding information, research facility, keywords, technical methods details and publication details. However, the biosample or plant accession metadata is not captured. Figshare recommends metadata submission that is similar to INSDC repositories but does not enforce it as a requirement. The storage quota for a free account is 20GB and up to 100 projects.

2.3.2. Genotyping data submission and metadata requirements

The major repository for submitting non-human VCF files containing genotyping related data is the European Molecular Biology Laboratory-European Bioinformatics Institute (EMBL-EBI) European Variation Archive (EVA) (26), but a newer repository has also arisen in the Genome Variation Map (GVM) (27). NCBI does host the dbSNP and dbVar databases, but those are intended for human data. All repositories strive to adhere to FAIR practices, but additional recommendations have been put forth by others (28). The EVA repository accepts VCF file structures, including hapmap formatted files (29) and SNP genotyping arrays, that are validated through a custom EBI VCF Validation Suite software (https://github.com/EBIvariation/vcf-validator) with a minimum number of data fields with accompanying metadata that includes, but are not limited to, project title, sequencing platform information, software, reference organism and genome version, and date and data generation location. The data fields for a VCF are the header lines that contain information about the dataset and relevant reference sources (organism, genome version, alignment and mapping method, etc.) followed by the variant site record row data: chromosome number, chromosome position, reference allele, alternate allele, quality, filter tag, and additional allele info format (https://gatk.broadinstitute.org). However, the naming structure within some of these fields is not standardized, which can lead to interoperability concerns.

2.3.3. Crop/clad Community GGB Databases

Whole genome, transcriptome, and genotype data can also be submitted to most of the GGB databases such as GDR (30-32), CottonGen (33,34), SoyBase (35,36), LIS (37,38), SGN (39,40), MaizeGDB (41,42), TreeGenes (43,44), TAIR (45,46), KnowPulse (47), and InterMine (48,49) databases (Table 2). Some of these databases, such as Gramene (50-52), SorghumBase (53) and InterMine (48,(54), do not accept data from authors but obtain from the primary databases. Depending on the GGB databases, different types of data and metadata can be submitted. Typically, these crop GGB databases collect a wide variety of data such as QTL, GWAS, markers, alleles, genetic maps, and cultivar/germplasm phenotyping data, and integrate them with whole genome, transcriptome, and genotyping data. These GGB databases standardize various names, associate the data with various ontologies to integrate data from various sources and of various types. This integration of different types of data, not typically done in the primary databases specialized in particular types of data, is one of the key steps in making the data FAIR. Integrating data from diverse sources enables researchers to discover novel associations between different types of data, potentially leading to valuable insights and breakthroughs. For example, combining SNP genotype data and phenotype data from multiple locations of the same germplasm can reveal how particular genotypes manifest specific phenotypes in distinct environments. However, achieving such insights requires meticulous integration of data from multiple sources.

2.3.4. Uses and Applications

WGS data can be reused in genome assembly, pan-genome construction, single nucleotide variation (SNV), copy-number variation (CNV), and structure variation (SV) discovery, phylogenomics, comparative genomics, and other genome research to study genome structure, genome diversity, the evolution of gene families or organisms, and crop domestications. Genotyping data in VCF format can be used for numerous purposes: storage of the location of given variants (including GWAS-associated variants); to identify targets of molecular markers for genotyping purposes; evaluating the effects of given base pair and structural variants on gene function; comparative genomics and evolutionary studies; and computational breeding approaches via machine learning and other methods. Data extraction and manipulation of VCF files is easy with the use of existing software toolkits such as VCFtools and SAMtools and can be utilized in conjunction with existing and ad hoc bioinformatic pipelines due to its command line functionality. By integrating VCF data with RNA-Seq and phenomics data, researchers can utilize these data sets quantitative genetic studies including genome-wide association studies (GWAS), quantitative trait loci (QTL) analysis, marker discovery, and genome selection (GS), to accelerate modern breeding techniques. Integrating transcriptomics data with metabolomics data can help in predicting biomarkers, which are often associated with biological pathways. This will assist in understanding the mechanism of underlying molecular patterns driving a condition. Integration of genomic, epigenomic and transcriptomic profiles will facilitate the prediction of key genomic variables and biological variation. Integration of gene expression data and copy number variations can be used to categorize samples into groups based on their similarity to two datasets.

3. Phenotype and Phenomics

3.1. Data types, Repositories, and Knowledge Bases

Plant phenotyping is the key for plant breeding, characterization of biodiversity, and genetic and genomic-based approaches for translational research (55). The classical genetic and functional genomics studies in model and crop plants have identified numerous mutants that show distinct morphological and anatomical mutants and associated the individual mutant phenotypes with one or more genes, pathways and molecular processes. Table 3 lists databases that host the mutant collections and description of phenotype of individual mutants and associated genes, including MaizeDIG (42), RIKEN Arabidopsis Genome Encyclopedia (56), Mutant Variety Database (57), Plant Genome Editing Database (58), Tomato mutant Archive TOMATOMA (59), and Plant Editosome Database (60).

In addition, complex phenotypic traits (i.e., morphological and physiological) related to the fitness and performance of an organism are often quantitative in nature and have multiple genetic determinants (61,62). Examples of traits that are determined by multiple genes (known as Quantitative Trait Loci, QTLs) are crop yield, biomass, resistance to pests and pathogens, abiotic stress tolerance, nutritional value, and ease of harvest. In addition to crop breeding, trait-based approaches are widespread in ecological research (63), as they provide a general understanding of a wide range of ecological and evolutionary phenomena such as impact of climate change, and anthropogenic land use on biodiversity (64-66). In Table 3, we provide a list of a key databases (or portal of bigger databases) that host information related to traits, QTLs, and associated data including the Gramene QTL database (67), QTL database for wheat (68), GLOPNET (69), TRY (70), a database of Ecological Flora of the Britain and Ireland (71), BIOPOP (72), GRIN (73), the USDA PLANTS Database, BiolFlor (74), LEDA Traitbase (75), BROT database of plant traits for Mediterranean basin species (76), and AusTraits (77). Trait and QTL data are also integrated with other types of data in various crop community databases listed in Table 2.

Phenomics is the systematic analysis for the refinement and characterization of phenotypes on a genome-wide scale. With the advent of high-throughput platforms, it became possible to collect phenomics data at a single cell, organismal and/or population-wide scale (78). Phenomics can be used for species recognition and biodiversity characterization (79), for stress quantification (79-81), and for crop yield prediction (82,83). Thus, phenomics data sets are very large and have different formats (e.g., JSON file). Some of the databases that host phenomics data include GnpIS (84,85), PGP (86), Cartograplant (87), AgData commons (https://data.nal.usda.gov/; (88), PathoPlant (89,90), PncStress (91), OSRGD (92).

Despite its analogy to genomes, it is not possible to fully characterize phenomes due to heterogeneity and multifaceted nature of phenotypic data with added layers reflecting complexities at the cell, tissue, and whole plant level that have further variations according to development stages, and growth environment (78,93). Thus, phenomics approaches may focus on specific factors of phenotypic data. For example, an intensive phenomics study may focus on high-throughput digital imaging across different stages and tissues of an organism under different growth stages or growth environments and may include quantitative data about plant height, biomass, flowering time, yield, and photosynthesis efficiency. Another study may employ orthomosaic images or time-series RGB images and remote sensing to monitor the algal blooms in the ocean (94). As phenomics data can be extremely variable in nature, necessary metadata includes information about plant species, tissue, developmental stage, environmental conditions, experimental design, data collection, processing, and analysis.

In addition to traditional phenotypes, molecular phenotypes include changes in the chromatin organization, transcripts, proteins, metabolites and ions (95-97). The quantitative changes in the gene expression, proteins and metabolite profiles in plants have far-reaching consequences for (i) the nutritional values of cereals, legumes, fruits, vegetables; (ii) the quality of bio products such as wine, beverages, vinegar, oil, and fuel; (iii) the ability of plants to adapt in response to various abiotic stress conditions; and (iv) the innate ability to defend against pests, pathogens, and herbivores (98-102).

Proteome and metabolome datasets allow the deeper understanding of an organism’s metabolic processes at the level of organ, tissue, and cell, as well as how these processes change in response to intrinsic developmental programs and environmental factors. Proteome datasets further confirm the subcellular localization, their comparative abundance between different tissues and cells, protein–protein interactions, and post translational modifications (103). Once the original proteomic datasets and associated metadata/manuscript have been submitted to public data repositories such as PRIDE (103-105), MassIVE (https://massive.ucsd.edu/ProteoSAFe/static/massive.jsp), JPOST (106,107), IProX (108,109), Panorama (110), and Peptide Atlas (111,112), they are made available for re-analysis and further exploration by other researchers. Metabolomics provides a comprehensive overview of the metabolite profile of an organism, tissues, cells, or subcellular component at a specific time point and is used to identify nutritional, medicinal, flavor, and disease resistance compounds as well as chemical interactions between plants and other biological systems. A recent comprehensive review of the methodologies to explore the highly complex and diverse metabolites of plants and associated methodologies can be found in Tsugawa et al., 2021 (113). The types of data collected for metabolomics depends on the method of chemical fingerprinting. As an example, in mass spectrometry (MS), a typical dataset would consist of a matrix containing information on the retention time and index (RT), mass-to-charge ratio (m/z), and peak characteristics such as the number and width. These data go through pre-processing which converts raw instrument data into organized formats using background subtraction, noise reduction, curve resolution, peak picking, peak thresholding, and spectral deconvolution. There are various software tools for analyzing metabolite data, each of which may be specific to a particular method of detection or instrument used in the analysis. The most popular software are MZmine, XCMS, MSdial, metaMS, Progenesis QI and MetAlign. For annotation for unknown metabolites, popular software tools include MS-FINDER, MetDNA, MetFamily, and GNPS among others. Raw file formats generated by the machines include d, raw, idb, cdf, wiff, scan, dat, cmp, cdf.cmp, lcd, abf, jpf, xps, mgf. Derived file formats are mzml, nmrml, mzxml, xml, mzdata, cef, cnx, peakml, xy, smp, scan. Due to the complexity of metabolomic data, several initiatives were undertaken. The Chemical Analyses Working group started the Metabolomics Standard Initiative (MSI) to develop metabolomic standards (114,115) with revisions suggested by (116). Community driven Metabolomics Society has a Data Standards Task Group focusing on metabolomics data standardization and sharing. This was followed by the Coordination of Standards in Metabolomics' (COSMOS) (117), and MetaboLights (118), for developing tools to ease submission of metabolomic data (119). ProteomeCentral and Omics DI serve as central repositories for these datasets, which are then re-used in protein knowledge bases (Uniprot and NeXtProt), genome browsers (Ensembl and UCSC), proteomics resources and other bioinformatics resources (ex. OpenProt and LNCipedia). The ProteomeXchange (PX) datasets are re-analyzed by different proteomics resources of the PX consortium, making data more reliable. The Paired Omics Data Platform (PoDP) (120) links the metabolomics data submitted to MassIVE or MetaboLights to genomes stored in NCBI or JGI. In Table 3 we list the two major repositories available for submission of raw and processed metabolome data, the NIH Common Fund's National Metabolomics Data Repository (NMDR) portal and the Metabolomics Workbench, and MetaboLights.

Some gene expression and metabolic phenotype often culminate in visible phenotypes, which can be described using the Plant Ontology terms (121-123). More recently, Plant Ontology terms have been extended to large scale phenomics data from a single species (124) to support the comparative phenomics in plants (125) and describe trait phenotypes expressed under specific developmental stage or specific environment and stress (126). For covering the genotype-phenotype gap, we need integration of multiple types of data including genotypic, large-scale phenome, gene expression, proteome and metabolome data, described using defined and standardized ontologies.

After collecting and generating phenotypic and phenomics data, it is recommended that they are formatted using community guidelines and submitted to primary data repositories, along with well-described metadata. The primary repositories serve as a source of primary or raw data (with base annotations) to the secondary databases for their visualization on genome browser (127) or for synthesizing new information by integrating them to other data types like plant metabolic networks (128,129), system-level plant pathways (130-132), expression Atlas, metabolic models, etc. These secondary knowledge bases are of primary importance to the plant researchers for formulating data-driven hypothesis for experimental and translational research and for analyzing the high-throughput omics data in the overall context of a species genome, systems-level pathway networks (133), and for gaining evolutionary insights by conducting intraspecies and interspecies comparisons. The implementation of standards and the development of infrastructure of public repositories are crucial for FAIR phenotypic data, even if many public repositories are currently not supporting the submission of the phenotype data (see Table 3).

3.2. Phenotype data formats, standards and metadata

The structure and characteristics of data types, along with any additional metadata, is crucial for enabling future data re-use and re-analysis by other researchers. The most relevant metadata shared across the various data types (generated by a diverse set of methods and platform) include taxonomic identification of the plant, the individual or cultivar name or accession ID, geo-references or growth conditions, field sampling or experimental design, cell, tissue, organ information (e.g., whole plant, leaf, root, flower, shoot, single cell, etc.), plant maturity and health status, measurement date (season, time of the day), and the type of phenotype measured (quantitative or qualitative) (70,134). These metadata can be entered as simple text format during the submission of the raw data to any primary repository and are easily exported from one database to another as TXT files.

Furthermore, plant phenotype/traits can be classified as categorical (qualitative and ordinal) or quantitative (continuous) traits (135). Some phenotypes are rather stable within species (mostly categorical traits), and some of these can be systematically compiled from species checklists and floras (e.g., (136). Thus, not all phenotypes can be mapped from one species to another. It is also important to note here that often, a phenotype is a cumulative outcome of the genotype, the environment and their interaction. Many important agronomic traits, such as seed or fruit quality, yield, abiotic stress tolerance, and pathogen resistance have a quantitative genetic architecture, involving minor and major genes or QTLs. Thus, the research question and the method become important to set the scope and goals of the study and require specific metadata and standards. For instance, most traits relevant to ecology and earth system sciences are characterized by intraspecific variability and trait–environment relationships (mostly quantitative traits). These traits have to be measured on individual plants in their particular environmental context. Each such trait measurement has high information content as it captures the specific response of a given genome to the prevailing environmental conditions (70). Thus, the collection of these quantitative traits and their essential environmental covariates is of vital importance. While trait measurements themselves may be relatively simple, the selection of the adequate entity (e.g., a representative plant in a community, or a representative leaf on a tree) and obtaining the relevant ancillary data (taxonomic identification, soil and climate properties, disturbance history, etc.) may require sophisticated instruments and a high degree of expertise and experience. Besides, these data are most often individual measurements with a low degree of automation. This not only limits the number of measurements but also causes a high risk of errors, which need to be corrected a posteriori, requiring substantial human work. Hence, the integration of these data from different sources into a consistent data set requires a carefully designed workflow with sufficient data quality assurance. These measurements of quantitative traits are single sampling events for particular individuals at certain locations and times, which preserve relevant information on intraspecific variation and provide the necessary detail to address questions at the level of populations or communities (134). Hence, an accurate and careful collection of data, their associated meta-data and ancillary data, is key to correctly preserve this valuable information, as well as to perform a suitable data integration across studies, species and data types.

4. Association mapping (GWAS) and linkage mapping (QTL)

Genome-Wide Association Study (GWAS) and QTL mapping are statistical methods used to identify marker-trait associations and candidate genes (causative mutations) controlling traits of interest. Both approaches rely on the linkage disequilibrium (LD) between the tested markers and the functional polymorphisms at the causative genes. However, they differ on the type of genetic populations used for the study: GWAS relies on diversity panel (e.g., germplasm collections) of, ideally, unrelated individuals; on the contrary, QTL mapping investigates the co-segregation of genetic markers with desired phenotypes in progeny purposely generated (e.g., F₂ population or recombinant inbred lines). Regardless of the fact that they are both analytical methods, their results can be used as data inputs for other types of analysis (e.g. meta-analysis, estimation of polygenic scores) (137). The genomic and genetic positions of trait-associated markers from GWAS and QTL studies can also be integrated with other types of data, enabling data transfer among related species. Thus, their outputs can be considered as a data type, and consequently, they require metadata collection and the use of standards in order to make them FAIR. Therefore, the FAIRness of the association mapping outputs is also key to contributing to link genotype and phenotype in the multi-omics era.

The primary output of a GWAS analysis is a list of variant positions, SNP ID or Indel positions, allele, strand information, effect size and associated standard error, p-value and corrected p-value, test statistics, minor allele frequency and sample size (138). One of the key metadata for GWAS/QTL data is the type of statistical method used to calculate and correct the p-values (GWAS/QTL). Regarding the SNPs, the most important metadata include the model species and the version of the reference genome against which these SNPs are mapped (see Genotype data section). The metadata required to make the traits interoperable and reusable is explained in the section lab/field traits. In the case of QTL analysis, a linkage map and pedigree information of the individuals, as well as the heritability of each SNP, is also important to be collected (139).

Unlike the human and animal GWAS and QTL data open access resources such as the NHGRI-EBI GWAS Catalog, GWAS Atlas (140), OpenGWAS, Animal QTL database, and Animal Genome Informatics resources (USDA national infrastructure NRSP-8: A National Animal Genome Research Program), QTL and GWAS data for plant species and major crops are mostly stored in crop community database (Table 2). The databases typically integrate the QTL and GWAS data with other types of data, playing a crucial role in improving the findability and accessibility of plant GWAS data that would have otherwise been buried in publications. AraGWAS Catalog (140) contains recomputed GWAS results using a standardized GWAS pipeline on all publicly available phenotypes from AraPheno (141).

Meta-analysis is the widely used analysis for integrating the summary statistics from multiple GWAS/QTL studies. It is a set of methods that allows the quantitative combination of data from multiple studies, and the evaluation of the consistency, inconsistency, or heterogeneity of the results across multiple datasets. Meta-analysis of GWAS/QTL datasets can improve the power to detect association signals by increasing sample size and by examining more variants throughout the genome than each dataset alone (142). However, in order to integrate datasets coming from different studies in meta-analysis, a standardized data and meta-data collection among the studies is needed. In addition, the genotype and phenotype data from the GWAS/QTL studies can be reused for further knowledge discovery, especially for QTL by environment interaction, predicting plant response in new environments, linking genomes to complex phenotypes across species.

5. Data reusability limitations and challenges

Accessing, reusing and integrating analytic data from various data types remain difficult (143). Despite the significant progress made in agricultural research due to advances in genotyping and phenotyping technologies, most of the data used and generated in research studies are not shared. Even if the data is submitted to public repositories, it often remains inaccessible or unusable due to missing fundamental metadata or improper formatting. The lack of community-based guidelines of data sharing, coupled with the complexity of data size and type generated in GGB research are likely the biggest limitations for data reuse in this field. Another matter arises from various levels of attitude of research journals/editorials for data sharing requirements and policy. Here we discuss limitations to data reuse in genotype-to-phenotype studies in three aspects.

5.1. Challenges

Data diversity and data format heterogeneity. Agriculture and horticulture research involves a wide range of genotypic, phenotypic, and environmental data, which often come from different experimental protocols and data generation technologies, and data processing workflows. As a result, data formats can be highly heterogeneous, making it difficult to integrate data from different sources and reuse in future studies (70). This issue is even more significant for phenotypic data, especially with the new emerging high-throughput phenotyping technologies. Digital imaging and remote sensing allow researchers to explore new levels of trait variability that were previously inaccessible using traditional and manual phenotyping methods. However, the large variety of data and metadata generated by these technologies can be highly variable in terms of file size, format, and content. The heterogeneity of data analysis pipeline also contributes to the complexity of standardization in phenomics.

Data size, quality and versioning. Most genomics, transcriptomics, epigenetics, and phenomic data are extremely large in file size and computationally intensive. For example, whole-genome sequencing data used for variant calling or VCF files that collates multi-individual genome-wide variants can be computationally challenging to handle, limiting their sharing in FAIR public repositories, and making data manipulation difficult. Also, data quality and integrity may be compromised before or during the submission process, which can prevent their reuse.

Object identification. Data submitted to a public domain often lacks a unique data object identifier (e.g., DOI), and any plant or accession identifier (PID), which makes it challenging to trace and integrate different types of data generated from the same individual plant across experiments and research laboratories. To improve data findability and reuse, it would be desirable to have a universal DOI associated with its PID. However, most data used and generated in research studies are not shared or are inaccessible or not reusable because of missing fundamental metadata or improper format of the data.

Metadata and data standardization.

Metadata are any type of data descriptor that can facilitate data interpretation and reuse. It is very common that when data are submitted to public domains they are accompanied by incomplete, inconsistent, or missing metadata. Developing and promoting standard data formats and metadata can improve data discovery and reuse, facilitate data integration and interoperability, and allow data from different sources. Some data standards for genomics and phenomics data have been developed, such as the Minimum Information About a Genome Sequence (MIGS) from the Genomic Data Standards Consortium, the Plant Phenotype Ontology (PPO), the Minimum Information About a Plant Phenotyping Experiment (MIAPPE). For GWAS data, GWAS-VCF format (21) has been proposed. However, the promotion and consistent application of these standards across different research groups and databases remain a challenge. For instance, if there are standards for how to collect and describe trait measurements, they are organism-specific (e.g., International Organization of Vine and Wine (OIV); www.oiv.int) or based on model species.

The metabolomic research community faces similar challenges. An initiative to identify the grand challenges of metabolomic research was coordinated by the USA Plant, Algae and Microbial Metabolomics Research Coordination Network (PAMM-NET; (144)). As noted, the data obtained from metabolomic analyses can often result in different chemical features values even in the same biological treatments due to the variability associated with biological systems, differences in the equipment, differences in the protocol and reagents. Therefore, identifying metabolites with confidence and the limited metabolome depth of coverage are the key grand challenges in metabolomic research (144). A recent review of LC-MS literature found a lack of details reported on methodology and level of confidence for metabolites in most of the reviewed research articles (145). To address these challenges, multi-dimensional analyses methods, the use of standard libraries for metabolite characterization, and tools that simplify the submission of metadata and data are being developed (119,146).

Other barriers for FAIR data include considerations of data privacy and confidentiality, legal and ethical issues, concerns of ownership, lack of incentive if not credited for sharing data, lack of awareness of existing data standards or data repositories, or lack of resources to implement data standards.

5.2. Resources and Funding

The submission of different data types (i.e., genomics, transcriptomics, proteomics, metabolomics and phenomics data) to separate and specialized primary repositories is a common practice, resulting in a heterogeneity of data repositories and multiple PIDs, limiting data interoperability. It is challenging to locate phenotypic datasets for a particular set of plants that have been characterized at the genomic or transcriptomic level due to the absence of common standards among data repositories.

Incompatible software or hardware among different data platforms also make interoperability challenging. Bulk data download or data movement across repositories is another issue due to data size and a lack of standardization. Software development and maintenance are required for fast data search and retrieval, as well as sufficient user support. For some types of plant data such as QTL and GWAS, there are basically no primary repositories where researchers can submit their data. Community GGB databases (Table 2) addresses this issue by collecting, curating, and integrating various data types from different sources and related species, playing a key role in data integration. Not all plant GWAS data, however, are timely stored in databases due to either lack of crop community databases or funding for curation. In addition, community GGB databases often have limited computational and personnel resources for curation and inclusion of all types of omics data due to limited funding and lack of understanding of the importance of curation by funders. Additionally, there is a lack of appropriate infrastructure for the raw data deposition in community databases.

5.3. Implementation of FAIR data policy

FAIR data policy refers to the list of 15 guidelines elaborated to facilitate data search, access, and reuse by human-driven and machine-driven activities (3). These principles apply to every type of scholarly digital object archived in a repository, and their implementation has started in many different research fields (147-149). In summary, these principles recommend that, when data are submitted, they are very well described using richly detailed metadata, and are assigned a globally unique and persistent identifier that allows everybody to find them in a searchable resource. Data should be formatted according to community-based standards if available, or in a way that they can be easily interpreted and exchanged by human and computer systems. The use of controlled vocabularies and ontologies is strongly encouraged to facilitate data interoperability across database resources. FAIR data, however, do not mean open-access or free, but refers to clarity and transparency about the conditions governing access and reuse (e.g., credential system to access and download data; (150). All these principles together aim to increase data transparency and improve data reuse for new research purposes, enhancing data value across time.

The implementation of FAIR data policy, however, can be challenging due to several reasons. Firstly, making data FAIR requires additional efforts and time commitment from researchers, which can be a barrier to implementation. Secondly, many scientists are not aware of the FAIR principles, community-based standards, and ontologies available to make their data FAIR. Thirdly, there is also a need for long-term sustainability of database resources, which requires ongoing funding and infrastructure support. Many databases struggle to secure funding and may face difficulties in maintaining FAIR data quality and accessibility over time.

6. Recommendations

With the latest advances in DNA sequencing and phenotyping technologies, the analysis of large datasets can be used to study the genetic instructions from either a single gene or the whole genome to be translated into the full set of phenotypic traits of an organism. The phenotypic data must be high quality digital phenotypic data with robust metadata that can be used for further downstream analysis and mathematical modeling of the phenotypic traits such as development, stress tolerance etc. To improve the data collection and data sharing of genotypic to phenotypic data, here are the several recommendations that can be taken to ensure their interoperability and reproducibility.

• Standardization of data collection protocols: Standardizing data collection protocols and using common data formats can help to ensure that data is collected in a consistent and comparable way. Use of metadata standards and the requirement of new ontology terms will make it easier to share and compare data across different studies.
• Centralized data sharing platform: Developing and using centralized data sharing platforms, the use of standardized data models and exchange formats and the deployment of existing and emerging software components can help to facilitate the sharing of genotypic and phenotypic data among researchers. It includes the use of online databases and repositories that are specifically designed for the storage and sharing of the plant genetic and phenotypic data.
• Consistent data annotation: Consistently annotating data with relevant information such as the genotype, phenotype, and experimental treatments can help to make the data more easily searchable and usable by other researchers.
• Data quality control: More automated management of data flows and implementing data quality control such as data curation and validation can help to ensure that the data is accurate, reliable and can be used to make valid conclusions.
• Data integration: Adoption of new database technologies and the development of robust data standards can facilitate the global integration of G2P data in future. Data integration from different resources such as genomics, transcriptomics, proteomics and metabolomics can help to better understand the complex relationship between genotype and phenotype.
• Community driven efforts: Community driven efforts such as open-source projects, workshops and collaborations can help to promote the sharing and use of data among researchers, which in turn will lead to better understanding of the G2P relationship. There should be encouragement on integrated science training plans that enable biologists to think quantitatively and facilitate collaboration with experts in physical, computational and engineering sciences. It can help the scientists to get familiar with the development of computational pipelines and workflows that will be essential for researchers to acquire, analyze and critically interpret G2P data.
• Data storage infrastructure, data management software and data curation tools are necessary to handle the large volumes of data in diverse formats.
• A concerted effort to make multi-omics data sets interoperable by automated biocuration with controlled ontology terms will help address this issue. Community databases address some of this issue by collecting, curating, and integrating various data of different types, from different sources, and from different but related species. However, community databases need to have sustainable funding.
• Data security, backup and recovery must be considered and implemented for sustainability.
• Data compliance with data sharing policies, privacy regulations and laws should be enforced.