LSMA

LONDON SCHOOL OF MULTIOMICS & AI

Free course

Free course

0 of 73 lessons

Lesson 3 of 7

Module 2 · Biological data and databases

Identifiers and cross-references: following one gene across databases

Identifiers and cross-references: following one gene across databases

Identifiers and cross-references: following one gene across databases

Every database gives the same gene its own label, and its RNA and protein labels of their own. Learn to read those labels, follow the links between them, and avoid the traps of versions, genome builds and renamed genes.

⏱ 14 min readModule 2 · Biological data and databasesFree
Lesson at a glance6 steps · about 14 min
  1. One gene, many names
  2. TP53 across the databases
  3. Versions: the number after the dot
  4. Genome builds: GRCh37 and GRCh38
  5. When gene names turned into dates
  6. Mapping pitfalls

One gene, many names

The main databases of The main databases: sequences, genomes, proteins, structures were built by different teams, in different countries, for different jobs. Each one gives every record its own identifier: a short, unique label that points to exactly one record. So the same human gene ends up with a symbol, a number at NCBI, a code at Ensembl, and further labels for its RNA and protein.

There is a good reason for this. A database must be able to promise that its identifiers stay stable and never point to two things, and it can only promise that for labels it issues itself. So each one keeps its own system, and the systems are linked rather than merged. The cost falls on the user: a list of genes from one source rarely drops straight into a tool built around another.

The databases are joined by cross-references: links inside one record that point to the matching record in another database. A UniProt protein entry, for example, lists the Ensembl and RefSeq records it corresponds to. Following these links is how you move from a gene name in a paper to its sequence, its protein and its structure without losing your way. This lesson shows what each kind of identifier looks like, what it really points to, and where the traps are.

TP53 across the databases

We will follow one gene throughout: TP53, which makes the protein p53, a guardian against cancer that stops damaged cells from dividing. It is one of the most studied human genes, so every database has it.

In The main databases you met each resource and the kind of question it answers. Here the focus moves from the databases to their labels: what the TP53 record is called in each one, and how to tell those labels apart at a glance.

Identifier typeIssued byTP53 valueWhat it names
Gene symbolHGNCTP53The gene, by its approved short name
HGNC IDHGNCHGNC:11998The gene, by a number that never changes
Gene IDNCBI Gene7157The gene record at NCBI
Ensembl geneEnsemblENSG00000141510The gene record at Ensembl
Ensembl transcript and proteinEnsemblENST… and ENSP…, several of eachOne for each transcript, and one for each protein it makes
RefSeq mRNANCBI RefSeqNM_000546One curated transcript
RefSeq proteinNCBI RefSeqNP_000537The protein made from that transcript
UniProt accessionUniProtP04637The protein entry, with its function and features
LocationGenome assemblyChromosome 17, band 17p13.1Where the gene sits

How to read this table

Read each row as one kind of label. The first column says what kind it is, the second who hands it out, the third the value for TP53, and the last what the label actually points to. Notice that the rows fall into three groups: labels for the gene, labels for a transcript (an RNA copy of the gene), and labels for a protein. Mixing up those three levels causes most identifier mistakes.

  • HGNC is the HUGO Gene Nomenclature Committee, the group that approves one official symbol and full name for each human gene. Human symbols are written in capitals and in italics in papers (TP53); the mouse version of the same gene is written Trp53, with only the first letter capitalised. The HGNC ID is a plain number with an HGNC: prefix. Symbols are sometimes changed; the HGNC ID is not.
  • NCBI Gene is the gene database of the US National Center for Biotechnology Information. Its Gene ID is a bare number. It is easy to confuse with any other number in a spreadsheet.
  • Ensembl is the genome resource run by EMBL-EBI in the UK. Its human identifiers start with ENS, then a letter for the level (G for gene, T for transcript, P for protein, E for exon), then 11 digits. The letter alone tells you which level you are looking at.
  • RefSeq is NCBI's set of reference sequences. Its two-letter prefixes carry meaning: NM_ is a curated messenger RNA, NP_ the protein made from it, NR_ a curated non-coding RNA. Records starting with X, such as XM_ and XP_, are predicted: they were made by a computer pipeline from the genome and have not yet been reviewed by a curator. They are useful, but less trustworthy than the N versions.
  • UniProt is the main protein knowledgebase. Its accessions are six or ten characters, mixing letters and digits, such as P04637. Each entry also has a readable entry name, here P53_HUMAN, but the entry name can change, so cite the accession.
Identifiers and cross-references: following one gene across databases · LSMA free course One TP53 gene card, with HGNC:11998, Gene ID 7157 and ENSG00000141510, branches to three transcripts: the highlighted reference transcript (ENST and NM_000546), another coding transcript (ENST and NM_ or predicted XM_), and a non-coding transcript (ENST only). The two coding transcripts lead to proteins: ENSP, NP_000537 and UniProt P04637 for the first; ENSP, NP_ or XP_ and a UniProt isoform for the second. The non-coding transcript makes no protein. GENE TRANSCRIPTS PROTEINS TP53 HGNC:11998 · Gene ID 7157 ENSG00000141510 Reference transcript ENST… NM_000546 Other coding transcript ENST… NM_… or predicted XM_… Non-coding transcript ENST… makes no protein ENSP… NP_000537 UniProt P04637 ENSP… NP_… or XP_… a UniProt isoform No protein-level identifiers
Identifiers live at three levels. The gene has one set; each transcript has its own; each protein has its own again. "ENST…" stands for an Ensembl transcript identifier whose digits are not shown. The number and kind of transcripts here are simplified. Schematic.

How to read this figure

Start at the top and work down. One gene can be copied into several different transcripts, because cells can start at different points or splice the exons together in different combinations (Module 1). Each transcript gets its own identifier. Coding transcripts are then translated into proteins, which get identifiers of their own; a non-coding transcript stops at the RNA level. The highlighted path is the reference transcript: the one most databases and clinical reports use by default. Reading a gene record and a protein record explains how it is chosen.

The figure shows why "the TP53 identifier" is not a single thing. TP53 in fact makes several protein isoforms, slightly different versions of the protein made from different transcripts. UniProt keeps them in one entry, labelled with a dash and a number, such as P04637-2; RefSeq and Ensembl give each its own record.

Versions: the number after the dot

Sequence records change. A curator may correct a base, extend the end of a transcript, or redraw an exon. When the sequence changes, the accession stays the same but its version, the number after the dot, goes up: NM_000546.1, then .2, and so on. Ensembl does the same with its stable IDs, so a gene appears as ENSG00000141510 followed by a dot and a version. UniProt tracks sequence versions in the entry history instead of in the accession.

If you search for an accession without a version, most databases show the latest one, which may not be the sequence a report or paper used. Older versions are not deleted; the record's history page lists them with their dates. The version numbers of different databases are independent: version 4 of a RefSeq transcript and version 18 of an Ensembl gene say nothing about each other.

Why should you care? Positions are counted along a particular sequence. A variant is usually written in HGVS notation, the international standard for describing variants, in a form such as NM_000546.n:c. followed by a position and a change, where c. means "counted along the coding sequence". If an older version had a few extra bases at the start, every position along it is shifted, and the same text can point to a different base. That is why clinical reports always quote the transcript with its version, and why you should copy the whole identifier, dot and all.

Genome builds: GRCh37 and GRCh38

The human reference genome itself has versions, called builds or assemblies. The two you will meet are GRCh37, released in 2009, and GRCh38, released in 2013 and now the standard. The UCSC Genome Browser calls them hg19 and hg38. Between them, gaps were filled and errors fixed, so stretches of chromosome moved by small or large amounts.

The result is that the same variant, in the same person, has a different chromosome position in each build. A position is meaningless unless you know the build. Mixing builds in one analysis is a classic silent error: nothing crashes, but variants land on the wrong bases or the wrong genes. Converting positions from one build to the other is called lifting over, and dedicated tools do it. Gene identifiers such as the Gene ID or the ENSG number do not depend on the build, which is one more reason to join data on identifiers rather than on positions.

How do you tell which build a file uses? Start with the header. A VCF file often names its reference in the ## lines, or lists each chromosome with its length, and the chromosome lengths differ between builds. File names and the methods section of a paper may also say hg19 or GRCh38. If nothing says, ask whoever made the file rather than guess. A related pitfall: hg19 and GRCh37 hold the same chromosome sequences but name them differently (chr1 against 1) and differ in the mitochondrial sequence, so "the same build" can still need renaming before two files match.

Go deeper: patches, alternative versions and a new reference

A build is not frozen the moment it is released. GRCh38 has had a series of patch releases, the latest GRCh38.p14 in 2022. Patches add corrected or extra sequence alongside the chromosomes without moving any existing position, so a GRCh38 coordinate stays valid. GRCh38 also includes alternative versions of some highly variable regions, such as the immune-system genes on chromosome 6, because one sequence describes them poorly. Many pipelines leave these out, and files may differ in whether they include them.

Lifting over is not always possible. A position in GRCh37 can sit in a stretch that was removed, duplicated or reversed in GRCh38, and then it has no single partner. Conversion tools report these as failures; check how many before trusting the converted file. A third reference, the gap-free T2T-CHM13 from 2022, is used more and more in research, so expect to see it named too.

When gene names turned into dates

Gene symbols were designed to be read by people, and that has a cost. Spreadsheet programs try to be helpful: type SEPT1 or MARCH1 into a cell and many will silently turn it into a date, 1 September or 1 March. Surveys of supplementary files published with papers found such errors in a striking share of them.

The problem was common enough that HGNC renamed the affected genes around 2020. The septin genes became SEPTIN1 and so on, and the MARCH genes became MARCHF1 and so on. HGNC now avoids approving new symbols that a spreadsheet would misread. The practical lessons: format identifier columns as text before pasting, keep a stable ID such as the HGNC ID or Gene ID alongside the symbol, and be aware that older papers and files still use the old names.

Dates are not the only conversion. Spreadsheets can also turn long identifiers made only of digits and a letter E into numbers in scientific notation, and drop the leading zeros from codes that start with 0. The safest habit is to open data files with each column imported as text, or to work in a tool that does not guess at types, which is one reason analysts move to the command line and scripts for this kind of work.

The size of the problem was measured by Ziemann, Eren and El-Osta (2016, Genome Biology), who found gene-name errors in about a fifth of papers with gene lists in spreadsheet supplements. A follow-up by Abeysooriya and colleagues (2021, PLOS Computational Biology) found the problem had not gone away. HGNC's current rules, including the renaming, are set out in Bruford and colleagues (2020, Nature Genetics).

The share applies to the journals and years those studies screened, not to all papers. Also note that renaming fixes new files only: older supplementary tables keep their converted dates, so check any list you reuse.

Mapping pitfalls

Mapping means converting a list of identifiers of one type into another type, for example from symbols to Ensembl gene IDs. It looks mechanical, but four things regularly go wrong.

One gene, many transcripts
  • A gene ID maps to several ENST or NM_ records
  • Pick one on purpose, usually the reference transcript
  • Otherwise rows are duplicated or chosen at random
Aliases
  • Old or informal names, such as p53 or LFS1 for TP53
  • One alias can belong to more than one gene
  • Search the approved symbol, or check the alias list
Retired and merged IDs
  • Records are sometimes withdrawn or merged
  • An old ID may map to nothing, or to a new ID
  • Check the database's history or ID-conversion page
Wrong level or species
  • A protein ID is not a gene ID
  • TP53 is human; Trp53 is mouse
  • Match the level and species before joining tables

An alias is any other name a gene has been known by: a previous official symbol, or a name used in the older literature. Aliases are the reason a search for a short name can return several unrelated genes. Searching HGNC first shows the approved symbol, previous symbols and aliases on one page.

Two habits make any mapping safer. First, count what goes in and what comes out: if 500 symbols become 480 Gene IDs, find out which 20 failed and why, rather than letting them drop silently. If they become 530, some symbols matched more than one gene and need a choice. Second, do not mistake moving between species for mapping. Turning a mouse gene into "its" human gene is a question of orthology, which genes in two species descend from one ancestral gene, and it needs an orthology resource such as Ensembl's comparative genomics tables, not an identifier converter.

Worked example · a mixed identifier list

A colleague sends you a list of seven genes of interest, copied from several sources. You need one clean column of NCBI Gene IDs.

TP53 · 1080 · ENSG00000012048 · NP_000537
HGNC:1884 · P38398 · NM_000492

Step by step. First, sort each item by its shape, because the format tells you the database and the level.

TP53: short name, no prefix or underscore → gene symbol
1080: bare number → NCBI Gene ID
ENSG00000012048: ENS + G → Ensembl gene
NP_000537: NP_ → RefSeq protein
HGNC:1884: HGNC prefix → HGNC ID
P38398: six characters, letter first → UniProt protein
NM_000492: NM_ → RefSeq mRNA

Next, convert each one to a Gene ID, following cross-references up to the gene level. Searching NCBI Gene, or using an ID-conversion service, gives:

TP53 → 7157 · 1080 → 1080 (CFTR)
ENSG00000012048 → 672 (BRCA1) · NP_000537 → 7157 (TP53)
HGNC:1884 → 1080 (CFTR) · P38398 → 672 (BRCA1)
NM_000492 → 1080 (CFTR)
→ seven items, but only three genes: 7157, 1080, 672

What the answer means. The list looked like seven genes, but it held three genes written in different ways: TP53 twice, CFTR three times and BRCA1 twice. Without sorting the identifier types first, you would have counted seven, and any later analysis, such as "how many of my genes are in this pathway?", would be inflated. Two items were not genes at all but a protein (NP_000537, P38398) and a transcript (NM_000492); they had to be walked up to their gene. Going the other way, from gene to transcript, you would have to choose which transcript, because one gene can have many.

Key point

Before you use an identifier, know three things about it: which database issued it, which level it names (gene, transcript or protein), and which version or genome build it belongs to.

A second example: a variant file from 2015 and one from last year both list a variant at "the same" position in a gene. If one file used GRCh37 and the other GRCh38, the matching position is a coincidence at best. Check the build in each file's header first, lift one over to match the other, and only then compare.

Further reading · Module 2HGNC help: how human gene symbols are approved and searched. · Ensembl stable IDs: what ENSG, ENST and ENSP identifiers mean. · About RefSeq: the NM_, NP_ and XM_ prefixes. · UniProt help on accession numbers. · All reading for this course

Check your understanding

1. A table you were sent has a column of identifiers that start with ENSP. What do these identify?

2. Two files list a variant at the same chromosome position, but one used GRCh37 and the other GRCh38. What can you conclude?

3. Why did HGNC rename genes such as SEPT1 and MARCH1?

Up next

Lesson 4

Reading a gene record and a protein record

Ask BioTutor

Create a free website with Framer, the website builder loved by startups, designers and agencies.