Database Identifier Primer

Introduction

To do new work in informatics, we must often combine information from multiple data sources that we feel are connected. However, even simple biological concepts -- like a protein sequence -- have many definitions and can be recorded, maintained, and labeled differently by different institutions. This primer discusses common database identifiers and their relationships to each other, to help a user decide how to choose connections.

Quick Reference Table

Database	Identifier	Format (Example)	Uniquely Defines	New One Created By	Significant Relationships	Locally Available As
EMBL / NCBI	Primary Accession	A##### (X51700) (Nucleotide) AA##### (XA51700) (Nucleotide) AAA##### (CAA51700) (Amino Acid)	A sequence submission. Primary Accession is most recent accession; preferred method of citing a sequence. Can be amino acid or nucleotide (which can be determined by id w/o lookup).	New accession on merge / split of previous entries (e.g. 2 reads discovered to be part of same seq.)	One : Many Accessions (history of split / merge) One : Many GIs (history of revisions) One : At Most One SwissProt Entries (protein lists it as source in dbXref)	`GenBank.gi2accession.accession SwissProt.dbXref... CDART.gi2nr.nucleotide_accession`
	Accessions		Past accessions are maintained to recall merge/split history. Impossible to differentiate primary accession w/o full GenBank.		One : Main Primary Accessions (if split) One (nucleotide) : One (protein) GI -- Fasta output on protein blast often gives Accession of nucleotide source sequence.
	GenInfo ID (GI)	# (2359834) (nucleotide or amino acid indistinguishable)	The most unique identifier of a sequence. Someone at some point input this as a sequence, no more.	New GI on ANY revision (author corrects, adds, splits, merges, etc.)	One : Many Almost Anything One : One protein sequence from translation (FASTA)	`GenBank.gi2accession.gi CDART.gi2nr.gi`
SwissProt	Name	AAAAA_BBBB A is species code B is gene code	Unique to entry. Conserved between versions.	Unchanged according to docs.	One : One SwissProt Entry One : One Sequence (in entry) One : Many GB accessions, nucleotide and protein (history?)	`SwissProt.entries.name SwissProt.entry_set.name SwissProt.dbXref.name`
	Primary Accession	A#####	One per current SwissProt (protein) entry. Might change with version. Preferred method of citation. Used as primary ID in our DB -- change?	New accession for merge / split , or in order to group entries discovered to be similar.	One : One Name One : Many Accessions (history of split / merge) NO RELATION TO NCBI/EMBL ACCESSION.	`SwissProt.entries.accession SwissProt.dbXref.accession`
	Accessions	A#####	May indicate past SwissProt entry, retained in list between versions.		Each was once primary accession.	`SwissProt.entries.accession_list`
CDART	Protein Identity Group (pig)	# (6134)	Merges GenBank protein GIs with IDENTICAL sequence (different sources, different versions, different organisms) for sequence analysis.	New PIG on any difference in sequence.	One : Many GIs (identical) One : One GI (pig # = gi # of one member) One : One Nucleotide Accession (if given in DB, based on fasta output) One : At most One SwissProt entry (matching seqs) One : One Sequence (but it's unknown!!!!)	`CDART.gi2nr.pig CDART.hits.pig CDART.pig_set_membership.pig`

The Databases

GenBank

The authoritative collection of publically available sequence data. Almost any piece of data in any other database has links back to a GenBank record for the source sequence.

Identifiers

GenInfo Identifier (GI)
A GI is the most unique identifier to a sequence. Even if an author updates his sequence submission, his original sequence will remain accessible under the old GI. GIs are being replaced by version numbers appended to accessions. Amino acid sequences and nucleotide sequences both get GIs, and the two are indistinguishable.
Accession
A GenBank accession is intended to follow a piece of data through it's lifetime. If the data is revised or completed, the accession will not change. However, occaisionally splitting or merging entries will force the creation of new Accessions. Both Protein and Amino Acid sequences have accessions, but they are distinguishable by format.
TaxID
This is not a sequence identifier, but is significant for looking up sequences. Each TaxID represents a node on a tree that incorporates A) phylogeny data and B) symantec data. Some nodes may stand for families, others for the scientific name of a species, others for the common name of the species. Frequently, nodes at the species level are referenced to indicate the source of a sequence.
Unfortunately, multiple nodes can be used to represent multiple terms for the same species. As a result, referencers must either list all in their entry or pick one as their identifier. A querier trying to retrieve sequences by species must make the same decision. For example, in the lab we frequently use '10090' to stand in for mouse, even though the node it represents refers to "LK3 Transgenic Mice".

SwissProt

SwissProt focusses solely on being a protein database. Entries do not represent a sequence, they represent the protein itself -- including all reference papers and structural and functional annotations. The sequence is only one aspect of the entry. It is certainly not as complete as GenBank, but it has several advantages:

Selectivity -- SwissProt only admits protein sequences that are believed to be complete. In GenBank, many sequences may not represent transcribed material, and they may certainly not indicate the bounds of a studied protein.
NonRedundancy -- Because SwissProt is built around proteins not sequences, only one of each sequence is admitted.

Identifiers

Name
The unique reference for each SwissProt entry is the entry name. It is (at most) 10 characters long, in the format XXXXX_YYYY, where X is a 5 letter code for the species, and Y is a 4 letter code for the protein. Although both codes are helpful, they are not authoritative. The species name may mask the fact that the protein occurs in multiple species (use GB_taxID_list!). The protein name is not formalized, and the namespaces for these names are different by species (an identical one does not mean a homologue protein). The data in an entry may change with the version without changing the name.
Accession
Each SP entry contains multiple accessions. The first is the Primary Accession. It is uniquely correlated to the name, and is the preferred reference for a SwissProt entry. Unlike names, accessions are preserved from SP version to version in the accession_list. The reasons for accession changes are merges and splits of SwissProt entries. Therefore, each accession in the accession list was once the primary accession for a slightly different protein.

Conserved Domain Architecture Retrieval Tool (CDART)

As the name implies, CDART exists to display domain architectures for a large protein set -- all of GenBank, in fact. However, since we are reverse engineering the architecture annotation process, we use a few other aspects of its source data.

Identifiers

Protein Identity Group (pig)
CDART does not actually store sequences, but to simplify its task it stores correlations between sequences. GenBank may record many identical copies of a sequence, from different sources or versions. The table gi2nr correlates these into PIGs -- sets of GIs for which the sequence is exactly similar. Since CDART is going to be doing work only on sequences, all of it's annotation for a GI can be found by looking under the corresponding PIG.
Position Specific Scoring Matrix (pssmID)
CDART calculated architectures using models of domains from several domain databases (SMART, PFAM, CDD, COG, etc.), but our DB doesn't actually store those models. We do have a pssmID associated with each such model, though, which can be used to find hits of a given model on a given pig.
Cluster (clusterID)
This is our addition to CDART. We've imported data CDART generated indicating association between PSSMs -- as measured by their frequent overlapping co-occurrance in proteins. The ClusterIDs are given to a group of PSSMs that are connected to each other by this relationship.
This is intended to reduce the redundancy of the result set: ordinarily, a slightly different model for the same domain might exist in each of the domain databases CDART used. These cousin-PSSMs would be linked by the abover measurement, allowing us to group the results by cluster rather than PSSM. Ideally, these Clusters will better represent a domain than any one pssm, but this may or may not be the case.

back home
bobsedge@andrew.cmu.edu