Main

A wealth of genomic and metagenomic sequences

By the end of next year, there will be complete genome sequences of at least draft quality for more than 1,000 bacteria and archaea and 100 eukaryotes1,2 and for even larger numbers of viruses, organelles and plasmids. With the rapid pace at which new genome sequences are appearing, the need to consider how best to ensure stewardship of these data for the long term has never been more pressing.

Our genome collection: more than the sum of its parts. The analysis of genomic information is having an impact on every area of the life sciences and beyond. A genome sequence is a prerequisite to understanding the molecular basis of phenotype, how it evolves over time and how we can manipulate it to provide new solutions to critical problems. Such solutions include therapies and cures for disease, industrial products, approaches for biodegradation of xenobiotic compounds and renewable energy sources. With improvements in sequencing technologies, the growing interest in metagenomic approaches and the proven power of comparative analysis of groups of related genomes, we can envision the day when it will be commonplace to sequence tens to hundreds of genomes or more as part of a single study. At current rates of genome sequencing, it has been estimated that >4,000 bacterial genomes will be available soon after 2010 (ref. 1).

Given the importance of the growing genome collection, the capital investment in its creation and the benefits of leveraging its value through diverse comparative analyses, every effort should be made to describe it as accurately and comprehensively as possible. There is an increasing interest from the community in doing so, for three main reasons. The first is the interest in testing hypotheses about the features observed in genomes using comparative evo- and eco-genomic approaches3. The second is the need to supplement the content of a variety of databases with high-level descriptions of genomes that allow useful grouping, sorting and searching of the underlying data. The third is the growth in genome sequence data from environmental isolates and metagenomes—vast data sets of DNA fragments from environmental samples4,5,6. The data generated by such studies will dwarf current stores of genomic information, making improved descriptions of genomes even more important.

At present, both top-level descriptors and genome descriptions are incomplete for many reasons. First and foremost, in hindsight we now know the minimum quality and quantity of information that is required to make each description precise, accurate and useful. For example, even for bacterial and archaeal species with validly published names, strain names were not routinely captured in genome annotation documents before the sequencing of large numbers of genomes from the same species7, but such information is now considered essential. Through empirical observations, we are expanding our view of the types of information that are important for testing particular hypotheses, exploring new patterns and quantifying inherent sampling biases3,8.

As the number of habitats and communities sampled using metagenomic approaches increases, we are also being forced to rethink our understanding of the minimum information required to adequately describe a genome sequence. Without adequate description of the environmental context and the experimental methods used, such data sets will be of less value for researchers wishing to conduct comparative genomic studies or link genetic potential with the diversity and abundance of organisms. In fact, given the vast number of uncultivated microbes, it may be that a DNA-centric approach, in which genes are linked to habitats (locations), is more useful than the species-centric view9,10. Finally, sequencing technology is advancing rapidly, and the adoption of new methods11,13 will force the adoption of additional descriptors (e.g., the depth of sequence coverage, quality and whether any 'finishing' was used) to be able to distinguish among these methods.

Most often, metadata about genome sequences are found only in the primary literature or in reference works, such as Bergey's Manual14 for bacteria and archaea, rather than in sequence databases. The distributed and patchy nature of this information and the difficulties of curating even a few pieces of information for what are now very large collections of genomes make the vision of a single definitive source of rich genomic descriptions highly desirable.

The need for coordinated efforts

Facilitating and accelerating the process of collecting relevant metadata would clearly reduce ongoing replication of efforts and maximize the ability to share and integrate data within the genomics community. The obvious solution is to develop a consensus-based approach.

The Genomic Standards Consortium. The GSC is an open-membership, international working body formed in September 2005 (ref. 15). Its goal is to promote mechanisms that standardize the description of genomes and the exchange and integration of genomic data. The GSC community brings together (i) evolutionists, ecologists, molecular biologists and other researchers analyzing collections of genomes, (ii) bioinformaticians producing genomic databases, (iii) those who sequence genomes and (iv) computer scientists, ontology experts and members of other standardization initiatives, such as the International Nucleotide Sequence Database Collaboration (INSDC), which is responsible for the DNA Data Bank of Japan (DDBJ), European Molecular Biology Laboratory (EMBL) and GenBank databases (http://www.insdc.org/). The guidance of DDBJ, EMBL and GenBank will be critical to the success of the GSC initiative, both because they are the official stewards of the public collection of genomes and because of their interest in fulfilling community needs.

Minimum information about genomes and metagenomes

The GSC is working to define a set of core descriptors for genomes and metagenomes in the form of a MIGS specification (Fig. 1). MIGS extends the minimum information already captured by the INSDC. The MIGS checklist is given in Box 1, and the most up-to-date version is available from the consortium's website (http://gensc.sf.net). Examples of MIGS-compliant reports are given in Supplementary Table 1 online. The information required to comply with MIGS is routinely included in primary genome publications (or is referenced therein). However, this information needs to be formalized and made available in electronic form to improve its accessibility16 (Box 2).

Figure 1: The scope of MIGS.
figure 1

The MIGS specification enables description of the complete range of possible genomes (eukaryotes, bacteria, archaea, plasmids, viruses, organelles) and metagenomes. Core descriptors include information about the origins of the nucleic acid sequence (genome), its environment (latitude and longitude, date and time of sampling and habitat) and sequence processing (sequencing and assembly methods). MIGS-compliant reports can be rendered into an electronic format using the MIGS XML schema and controlled vocabularies through the GSC's Genome Catalogue (http://gensc.sf.net).

Since it was originally proposed16, the MIGS specification has been simplified and changed by the GSC through an iterative revision process to contain (i) only curated information that cannot be calculated from raw genomic sequence and (ii) core descriptors specific to the major taxonomic groups (eukaryotes, bacteria and archaea17, plasmids, viruses, organelles) and metagenomes. MIGS is structured as an 'Investigation' composed of a 'Study' and an 'Assay', according to the Reporting Structures for Biological Investigations (RSBI) working group's recommendation for the modularization of checklists18,19. Under 'Study' are the top-level concepts 'Environment' and 'Nucleic Acid Sequence' and under 'Assay' is a description of the sequencing technology.

MIGS aims to support unencumbered access to genomic reagents (such as strains)20, place the complete (meta)genome collection into geospatial and temporal context (latitude, longitude, altitude or depth, date and time of sampling) and provide essential details of the experimental method used (e.g., sequencing method). MIGS also provides a framework for the capture of extra information deemed 'minimum' to specific communities. Most importantly, the description of metagenomes in MIGS is being extended in the minimum information about a metagenome sequence (MIMS) specification21. MIMS enables the capture of further measurements that define habitat (such as temperature, salinity, pH, dissolved organic carbon) and extends the original structure of MIGS for describing a single (meta)genomic experiment to allow the capture of information from pooled samples and more than one independent sampling event (e.g., sampling along a transect4).

How genomes and metagenomes are described in public databases has evolved from how short, simple DNA sequences are described, without special attention to information such as the geographical origin of the sequence. Significant efforts are underway by the INSDC to adapt and extend the infrastructure for describing genomes through the Genome Project Metadata initiative22. The INSDC efforts are open to evolution, albeit at a conservative pace22, and it is the GSC's hope that much, if not all, of the MIGS specification will be included in the Genome Project Metadata initiative. A mapping between INSDC features and MIGS has been developed for the purpose of placing MIGS information into INSDC documents and is available on our website. Any fields that are not already formally defined by the INSDC Feature Table Document (http://www.insdc.org/files/documents/feature_table.html) can be represented within a structured comment block in INSDC records22.

A genome catalog

The development of any checklist must be an open and iterative process that involves a balanced group of participants. Moreover, mechanisms for achieving compliance are needed to facilitate widespread adoption of a checklist. Such mechanisms involve an appropriate reporting structure for capturing and exchanging data (file formats), software, databases and appropriate controlled vocabularies and/or ontologies for defining the terms used in the annotations. The GSC is working toward these combined goals and has created an online system for capturing MIGS-compliant reports (http://gensc.sf.net).

In brief, we have implemented the checklist as an XML schema and built a freely available Genome Catalogue system (GCat) (http://gensc.sf.net). GCat is designed to generate forms automatically and 'on the fly' from this schema for the sake of data input. It also allows users to view and search genome descriptions as they accumulate during the process of refining the MIGS checklist. The GCat system is generic and could be applied to the capture of more expressive metadata for subsets of genomes. Indeed, it is flexible enough to support the implementation of any checklist that can be structured as an appropriate XML schema (MIGS.xsd, being developed into the Genomic Contextual Data Markup Language (GCDML)). The GSC is also working in the area of controlled vocabulary and ontology development through the collation of controlled vocabularies already in use in the community and through contributions to the Ontology for Biomedical Investigations (OBI, previously known as the Functional Genomics Investigation Ontology (FuGO)23) and the Environment Ontology (EnvO) project (http://environmentontology.org). As a part of this process, GCat makes use of existing controlled vocabulary terms and accepts new terms.

Improving genomic databases

By design, MIGS contains only primary, curated information. This is because secondary, or derived, information that can be calculated from a genome sequence is subject to frequent change, can be generated using more than one method and should be acquired directly from those producing the calculations. Still, access to computed information (e.g., in the simplest cases, G+C content or total number of predicted proteins) should be made as easy as possible.

Genomic sequences and their initial annotations must be submitted to the INSDC (http://www.insdc.org/) (and subsequent high-quality, curated annotations derived from empirical observations to the Third Party Annotation data set24), but there are an ever increasing number of genomic databases containing a wide range of additional computations. Although GSC does not endorse any particular method of analysis or database, it supports increased transparency of such resources for the sake of accurate data interpretation and integration.

The first issue is that of exchanging calculated information. This could be facilitated in part by widespread adoption of a common exchange format, such as the Generic Feature Format Version 3 (GFF3) file format (http://song.sourceforge.net/gff3.shtml). There are many tools that support the reformatting of a variety of file types into GFF3, so database providers would find it straightforward to generate appropriate files. The availability of a wide suite of tools for downstream analyses of files in GFF3 format also means that users could combine the weight of evidence from many sources when examining a particular genome. This could reveal instances of systemic bias and therefore lead to better genomic annotations, as more composite features would be available and conflicting annotations could be highlighted for resolution.

Exchanging data also relies on common standards for computational analyses, and supporting data downloads is not enough, regardless of format. Data resources should also be expected, within reason, to provide clear specifications for how the data are generated (for example, standard operating procedures (SOPs) that describe computations such as gene prediction and operon and ortholog identification). One example of this type of documentation is provided in AboutIMG, a web-based description of the Integrated Microbial Genomes (IMG) system25.

In the future it should be far simpler to combine various genomic features, exact details of how they were generated and enough information about the provenance (origin) of the analyses to be able to transparently share data from different sources. Such interoperability, especially when provided by participating databases in a way that would enable automatic harvesting of the data (e.g., through web service technology), would multiply the individual value of these databases many times over and open up new opportunities to examine genome sequences in unprecedented detail.

Future directions

The effort required to achieve the degree of transparency advocated here is considerable but offers substantial and immediate benefits. We argue that the cost of achieving such standardization is trivial compared with the sums spent generating the data. The capture of MIGS-compliant information will not only facilitate comparative genomic and metagenomic analyses but also enhance the available descriptions of downstream '-omic' experiments based on genomic data. It will also enhance the much larger 'halos' of 16S ribosomal RNA sequences that are now available for many sequenced genomes and metagenomes. For example, the genome sequence of the marine bacterium Silicibacter pomeroyi26 is 'embedded' in a large number of environmental 16S rRNA sequences affiliated with the Roseobacter lineage, which is accompanied by a fairly extensive literature describing the distribution, ecology and other properties of this group27.

Through its ongoing efforts, the GSC hopes to stimulate discussion of the MIGS specification and solicit further feedback from the community. It therefore has an open call for participation and is eager to solicit MIGS-compliant genome reports (including batch uploads) and collect relevant controlled vocabulary terms useful in the description of genomes and metagenomes. GCat identifiers have been implemented and are available for past or future projects, and MIGS-compliant genome reports are starting to become available online (e.g., refs. 28, 29, 30, 31). We expect a production version of MIGS (2.0) to be released by early 2008 with an appropriate set of terms formalized within OBI19 and other relevant Open Biomedical Ontology (http://obofoundry.org/) ontologies. We would hope that this milestone (release of MIGS 2.0) will be accompanied by recognition by journals and implementation by a variety of databases. Beyond this, the MIGS specification should still remain flexible enough to allow it to be revised in accordance with advances in technology and our biological knowledge. It should also be considered for use in combination with other checklists in the context of the Minimum Information about a Biomedical or Biological Investigation (MIBBI) Foundry (http://mibbi.sf.net), of which the GSC is a founding community19. The most up-to-date information about GSC activities is available at our website (http://gensc.sf.net).

Note: Supplementary information is available on the Nature Biotechnology website.

Disclaimer

Opinions, findings and conclusions or recommendations expressed in this paper are those of the authors, and do not necessarily reflect the views of the US National Science Foundation.