Chromosomes

Complete chromosome sequence assembled from individual clones (that are themselves available from the INSDC) is propagated into a RefSeq record. For some genomes, the RefSeq representation uses a unit of interest to the research community; for example, some of the RefSeq genomic records for Drosophila melanogaster represent chromosome arms rather than complete chromosomes. RefSeq records may also be available for some genomes that are not yet fully sequenced but for which complete sequence is available for individual chromosomes. These complete chromosome RefSeq records may be annotated by the NCBI computational annotation pipeline, or they may be curated by an organism-specific collaborating group and undergo NCBI validation before being released.

Microbial genomes

For microbial species, historically all complete and draft genomes submitted to GenBank were propagated to the RefSeq collection. This is no longer tenable because of the volume of genomic data being generated, so additional RefSeq records are created from new GenBank submissions only to span the taxonomic diversity; this means in general, one genomic RefSeq per species is provided. If significant sequence diversity exists, or if subspecies or subgroups require representation as determined by NCBI staff, more than one RefSeq may exist for a given species.

Small complete genomes

RefSeq records representing organelle, viral, and plasmid genomes are based on single GenBank records. For organelle and viral genomes, if more than one GenBank submission is available for a species, typically only one is chosen to propagate to the RefSeq collection. Various factors, including the level of annotation, strain information, and community input are considered when deciding which GenBank submission to represent. There is no plasmid taxonomy; a GenBank submission is propagated to the RefSeq collection if it is part of a larger registered genome sequencing project, or if it exhibits significant sequence divergence when compared to other plasmids.

Targeted loci

The RefSeq Targeted Loci Project is a collaborative effort to curate and maintain molecular markers of use in the identification and classification of organisms. The initial focus is on ribosomal RNAs, although expansion to other informative sequences is anticipated. From GenBank submissions, the project creates RefSeq records for the small subunit of ribosomal RNA (16S in prokaryotes and 18S in eukaryotes) and the large subunit ribosomal RNA (23S in prokaryotes and 28S in eukaryotes). As of November 2010, there are 3331 16S rDNA RefSeq records from bacteria and archaea and 137 18S rDNA, and 97 28S rDNA RefSeq records from fungi.

Curation of Microbial, Viral, and Mitochondrial RefSeqs

Microbial, viral, and metazoan mitochondrial RefSeq records are validated for content propagated from the original GenBank submission, including taxonomy, publications, and annotation, prior to becoming public. This content may be modified, augmented, or deleted by NCBI curation staff.

For microbial genomes, a set of minimal annotation standards (described here) are automatically provided on all legacy and new RefSeq records. These include ribosomal RNAs, transfer RNAs, and protein-coding genes with locus_tags. Ribosomal RNAs are predicted using BLASTn tools against an RNA sequence database and/or using Infernal (Eddy, 2002) and Rfam models (Griffiths-Jones, et al, 2003). Transfer RNAs are predicted using tRNAscan-SE (Lowe and Eddy, 1997). Other annotation above the minimum standards may be added based on an external source or literature review. Annotation associated with the NCBI’s Protein Clusters database is also propagated to the RefSeq records (both proteins and genes) at selected intervals. The Protein Clusters database is a collection of RefSeq proteins from complete genomes broadly organized into the following groups: archeal and bacterial genomes and plasmids, viruses, protists, plants, and chloroplasts and mitochondria, and annotated based on sequence similarity and protein function. This clustering allows the entire group to be curated as a single set, permitting well characterized proteins to seed the annotation of less studied ones within the same cluster. NCBI staff use literature and information from other databases, including UniProtKB/Swiss-Prot, to annotate each cluster with standardized protein names, biochemical descriptions, and other data, which is then transferred to individual proteins within the relevant RefSeq records. A microbial genome RefSeq record typically has a PROVISIONAL review status.

Annotation of viral genomes relies on an established group of Viral RefSeq Genome Advisors, members of the International Committee on the Taxonomy of Viruses, and other experts outside of NCBI. For example, the HIV-1 RefSeq (NC_001802) was curated by NCBI staff in collaboration with the authors of the book Retroviruses, and many of the adenovirus and herpesvirus records have been curated by outside experts. Based on literature review, NCBI curators may modify the CDS and RNA annotation compared to the GenBank submission, as was done for the Measles virus RefSeq record (NC_001498). Additional NCBI resources used during the curation of viral RefSeq records include the Protein Clusters database and PASC, a virus classification tool used to validate the taxonomy of virus RefSeq records across a number of taxonomic families. NCBI also maintains several specialized annotation pipelines for use in the Virus Variation and Influenza Virus resources. Manually curated viral RefSeq records are annotated with a status of REVIEWED or VALIDATED in the RefSeq COMMENT block.

For metazoan mitochondrial RefSeq records, standardized protein, gene, and RNA names are annotated independent of species-specific nomenclature guidelines. Additional curation may include adding common names or missing tRNAs and adjusting the coding region spans based on the Protein Clusters database. Curated metazoan mitochondrial records are annotated with a status of REVIEWED. Non-metazoan and plant chloroplast RefSeq records are not curated, are derived entirely from the original INSDC submission, and have a status of PROVISIONAL.

For targeted loci, vector or primer sequence from the GenBank submission is excluded from the RefSeq record. Any feature annotation may be modified to represent a standard format, and collection identifiers and publications referencing the original GenBank submission may be added.

Curation of Vertebrate and Invertebrate Records

Curation of higher eukaryotic organisms is focused on mammalian genomes, especially human and mouse, but also includes many other species with existing or planned genome assemblies. The RefSeq processing for these organisms provides transcripts and protein records as well as some genomic region records representing gene clusters or pseudogenes; these genomic region records facilitate genome-wide annotation. Because RefSeq uses evidence independent of a genome assembly to represent RNAs and proteins, the dataset can represent sequence not currently part of that genome assembly. RefSeq processing integrates the official nomenclature and other information, including alternate names, Gene Ontology (GO) terms, and literature and GeneRIFs available in Gene. Multiple collaborations support the collection of this descriptive information (Table 4; see also Chapter 19).

Sequences enter RefSeq curation processing by a combination of computational analysis, collaboration, and in-house curation. As illustrated in Figure 2, generation of the initial RefSeq record depends on identifying a representative sequence for a gene. New genes and sequence data are added to the in-house version of the Gene database by RefSeq curators, collaborators, NCBI’s genome annotation pipeline, and NCBI-based mining of UniGene, cDNA alignments, and INSDC submissions. Quality assessment (QA) processes are executed regularly to identify questionable data for review. These assessments include analysis of nomenclature, sequence similarity, genomic placement, and potential cloning errors (e.g., chimeras). The QA steps also leverage data from other NCBI resources, including HomoloGene, Map Viewer, and GenBank related sequences. Data conflicts must be resolved before the INSDC submission is used to generate a RefSeq record.

A sequence record unambiguously associated with a Gene record may be propagated into a RefSeq record. The completeness of the sequence (e.g., complete vs. partial CDS) and the category of the gene (e.g., protein coding, pseudogene) determine whether a RefSeq will be made, and if so, of what type (DNA, RNA, mRNA plus protein). RefSeq records are not made for incomplete proteins, transposable elements, or those loci for which the product type is uncertain (e.g., protein coding or not). It should be noted, however, that the RefSeq collection does include partial transcripts and proteins that are provided by collaborating groups or when the RefSeq is based on an annotated whole genome sequence submitted to the INSDC.

Once a suitable “source” sequence is identified, the RefSeq record is generated using the sequence data from the INSDC submission and the annotation data from the in-house version of the Gene database. Information from Gene includes the GeneID, cross-references to other databases, official nomenclature, aliases, alternate descriptive names, map location, and citations, including those submitted as GeneRIFs. RefSeq records are also subject to programmatic validation to identify annotation format errors and to provide annotation in a more consistent format. Records at this stage have a PROVISIONAL, PREDICTED, or INFERRED status depending on the evidence existing in support of the Gene record.

RefSeq processing for non-protein-coding RNA loci uses the longest defining transcript record associated with the Gene record. For non-transcribed loci (such as non-transcribed pseudogenes), the RefSeq record is typically derived from a region of a larger genomic sequence. Curation of these types of records is minimal because the current focus is on curation of protein-coding loci; however, these records provide an important reagent for the computational annotation pipeline and support annotation of non-protein-coding genes that might otherwise be missed or misrepresented as a predicted protein-coding gene.

Other RefSeq records are provided to represent larger genomic regions, including RefSeqGene sequences, gene clusters, genes requiring rearrangement to express a product (immunoglobulins and T-cell receptors), and haplotypes with known differences in gene content. These genomic region records are annotated by NCBI curation staff, often in collaboration with scientific experts, and are not provided by automatic processing.

RefSeqGene, a partner of the international Locus Reference Genomic (LRG) collaboration, provides stable reference standard genomic, RNA, and protein RefSeqs for medically important genes. These standards support the HGVS expressions used to describe sequence variation in medical records, and thus are constructed to represent standard alleles. The RefSeqGene usually represents a single gene, on the positive strand of the sequence, beginning 5 Kb upstream and extending 2 kb downstream. RefSeqGene records also include alignments of the RefSeq transcripts for the gene. All sequences annotated on the RefSeqGene have a review status of VALIDATED or REVIEWED.

Additional curation of vertebrate and some invertebrate RefSeq records occurs at the request of public users and collaborators, or as indicated by in-house QA analyses. QA analyses focus on, but are not restricted to, HomoloGene-based reporting of inconsistent protein lengths, identification of RefSeqs with repeat elements, questions about gene-to-sequence associations or potentially redundant genes, and reports of genes annotated at one time on a genome but not during subsequent re-annotation of that genome. Additionally, alignment-based tests are conducted for human and mouse that identify RefSeq records with poor quality alignment to the genome, non-consensus splicing, or very short or very long exons. Review of these records by skilled curators results in the most current and complete representation of the nucleotide and protein sequence and feature annotation available at that time. Sequence review may allow removal of vector and linker sequence, extension of the UTRs to define the full-length transcript, modification of the CDS annotation associated with the original INSDC source accession, or the creation of additional RefSeq records to represent the products of alternative splicing. A variety of feature annotations can be added to the RefSeq transcript and protein records. For nucleotide records, these include an indication of the transcript completeness, location of poly(A) signal and site, and sites of sequence variation and RNA editing. Exon annotation is provided for RefSeq transcripts and non-transcribed pseudogenes of human and mouse only; for transcripts, exon annotation is determined from the alignment of the transcript to the reference genome assembly using Splign, and, for non-transcribed pseudogenes, from the Splign alignment of the functional gene to the pseudogene genomic region. For protein records, feature annotations may include alternate or non-AUG initiating codons, Enzyme Commission (EC) numbers, mature peptide products, protein domains, and selenocysteine residues. Finally, literature review is another source of alternate names, aliases, and functional information, the latter which may be used to construct a Reference Sequence Summary on the RefSeq record. A RefSeq record that has undergone the complete review process has a REVIEWED status. Note that for many genes, intermediate levels of manual curation may address issues concerning the RefSeq sequence alone; these records have a review status of VALIDATED pending full review.

The review process may result in updating a RefSeq record, providing new RefSeq records, modifying sequence-to-gene associations, merging Gene records, or discontinuing a RefSeq, GeneID, or both. A RefSeq record is suppressed if it is found to represent a transcribed repeat element, to be derived from the wrong organism (i.e., the INSDC sequence it was based on has incorrect organism annotation), or not to represent a "gene". Records determined to represent an incomplete sequence, such as a partial protein sequence or an incompletely spliced transcript, are temporarily suppressed until more complete sequence data are available. Suppressed records can still be retrieved and will have a disclaimer appearing on the query result document summary (Figure 3a). A suppressed record is not included in BLAST databases, in the calculation of related sequences, in the BLink display (BLink are pre-computed protein BLAST results), or in RefSeq FTP releases. If a RefSeq is found to be redundant with another public RefSeq, then one is retained and the other becomes secondary (Figure 3b). If the sequences were associated with two different Gene records, then the records are merged so that a query of Gene with either of the original GeneIDs will retrieve the remaining single record.

Figure 3.

Suppressed or redundant RefSeq records. (A) A standard text statement is included on the Entrez document summary for suppressed RefSeq records. (A) If redundant RefSeq records are merged, then both accession numbers appear on the flat file ACCESSION line (more...)

We welcome input from the research community to improve the quality of the RefSeq collection. Interested parties are invited to contact us by sending an email to the NCBI Help Desk (vog.hin.mln.ibcn@ofni) or by using our feedback form.