Variants to meaning

In Reads to variants you turned NA12878's exome reads into 218 confident variant calls in the CYP2C region of chromosome 10, and found that she carries one copy of CYP2C19*2. A VCF only says where she differs from the reference. Today you will find out what those differences are – which gene, what they do to the protein, how common they are, what prediction tools say – and learn how laboratories sift thousands of variants down to the few that matter.

About 2 hoursNo software to installUses live data from Ensembl

The workbench

Variants – your calls in a table; the Annotate with VEP button sends their positions and alleles to Ensembl's Variant Effect Predictor and fills in the rest. IGV – NA12878's reads, as before. 3D viewer – crystal structures of the CYP2C enzymes.

buttons do a step for you – try it yourself first. Steps tick themselves when the page sees you do them; your answers are saved in this browser (My answers downloads them).

Your name (optional – it goes on your downloaded answers):

Where this fits

reads→alignment→variant calls→annotate→filter & prioritise→interpret

Two kinds of question are asked of variant lists, and they need opposite filters:

Pharmacogenomics

Which known variants does this person carry in genes that handle drugs? These are often common (CYP2C19*2 is on 1 in 7 European chromosomes). NA12878 is our example.

Rare-disease diagnosis

Which rare, damaging variant explains this patient's condition? Common variants are thrown away first. Your coursework patient is this kind of case.

Chapter	What you will do	Time
	Run Ensembl VEP on your variants	15 min
	Consequence terms, impact, transcripts, HGVS names	20 min
	Population frequencies in gnomAD	15 min
	SIFT, PolyPhen-2, CADD and SpliceAI	15 min
	How CYP2C192* breaks the enzyme	15 min
	Filter like a diagnostic lab; star alleles and drug response	20 min
	Map the variants onto CYP2C19, CYP2C9 and CYP2C8	20 min
	How to present a variant; quiz	10 min

Sections marked Extension are optional and not counted in your progress bar.

By the end you should be able to…

annotate variants with VEP on the right genome build and read its output: consequence, impact, transcript, HGVS;
use population frequency (gnomAD) and prediction scores (SIFT, PolyPhen-2, CADD, SpliceAI) as evidence – and know their limits;
build a filtering strategy that fits the question, and explain what it can miss;
show a variant's position in a protein structure and reason about its effect;
present variants in a clear, complete table for a report.

Annotating variants with VEP

≈ 15 minWorkbench: Variants

Annotation looks up each variant against gene models and databases. The Variant Effect Predictor (VEP) from Ensembl/EMBL-EBI is the most widely used annotator; diagnostic labs run it on every exome. You can use it through its web page, as a command-line program, or – as here – through its public web service.

Use the right genome build

Your variants have hg19/GRCh37 coordinates, so they must be annotated against GRCh37. Ensembl's main site uses GRCh38; the GRCh37 version lives at grch37.ensembl.org (this page uses its VEP service). Annotating hg19 positions against GRCh38 gives wrong genes and nonsense consequences – without any error message.

Activity 1.1 · Before annotation

Open the Variants tab. Your filtered calls are listed with position, alleles, QUAL, genotype and the reads for each allele – everything a VCF knows. (There are 219 rows for your 218 calls: at chr10:95,246,307 NA12878 carries two different alternative alleles, A>G and A>T, and each gets its own row.)

Activity 1.2 · Run VEP

Press Annotate with VEP. The page sends only chromosome, position and alleles – the same as pasting your VCF into the VEP web page – to Ensembl, in batches. It takes 10–60 seconds depending on how busy Ensembl is. If it fails or is slow, press Use saved results (the same annotations, saved when this practical was written).
Look at the new columns: Gene, Consequence, Impact, Protein change, the variant's database identifier (Known as), its population frequency (gnomAD AF) and four prediction scores.
Click the Impact column heading to sort by it (click again to reverse).

Which genes do the variants fall in or near? (Use the Gene column; sort by it.)

The region, chr10:95–98 Mb, holds far more than the CYP2C cluster: about 30 genes appear. The most variants are in MYOF (28), ALDH18A1 (17), SORBS1 (16), PDE6C (14), PDLIM1 and BLNK (13 each); the pharmacogenes have CYP2C9 9, CYP2C18 9, CYP2C8 8 and CYP2C19 6. Some are in non-coding RNA genes or pseudogenes (names like RP11-… or CYP2C58P), and 7 lie between genes.

How many of the variants are not already in dbSNP (“novel”)?

Two of 219 – both small insertions/deletions in repetitive sequence outside the exons, where calls are the least reliable. Everything else is already catalogued: most of anyone's variants are common and have been seen many times (and NA12878 herself has been sequenced by many projects). In a typical exome only a small percentage of variants are novel.

Consequences, transcripts and names

≈ 20 minWorkbench: Variants

VEP describes what a variant does with standard Sequence Ontology terms, and groups them into four impact classes. The terms are defined per transcript: the same variant can be missense in one transcript and intronic in another.

Impact	Typical consequences	Meaning
HIGH	stop gained, frameshift, splice donor/acceptor (±1–2 bp), start lost, stop lost	Likely to destroy the protein (loss of function)
MODERATE	missense, in-frame insertion or deletion	Changes the protein sequence; effect uncertain
LOW	synonymous, splice region (3–8 bp into the intron, 1–3 bp into the exon)	Probably harmless – but not always (chapter 5)
MODIFIER	intron, 5′/3′ UTR, upstream, downstream, intergenic	Outside the coding sequence; effect hard to predict

Activity 2.1 · What kinds of variant are there?

Untick LOW and MODIFIER in the Impact filter to keep the variants that change the protein.
Read the count above the table, then tick the boxes again.

How many of the variants have HIGH or MODERATE impact?

12 – all missense; there are no HIGH-impact variants at all. They are in CEP55 (2), PDE6C, FRA10AC1 (2), NOC3L (2), CYP2C18, CYP2C9, CYP2C8 (2) and SORBS1. The other 207 have LOW impact (20: synonymous and splice-region) or are MODIFIERs (187, mostly intronic).

Most variants are MODIFIERs. Why, when these reads came from an exome capture?

Capture probes pull down the exons, but the captured fragments are longer than the exons and some off-target DNA comes along too, so reads also cover the flanking introns, UTRs and nearby intergenic DNA. And across the genome, variants in introns and between genes far outnumber coding variants. A whole exome typically has about 20,000 variants in coding exons – roughly half synonymous and half changing the protein (the table in the lecture) – and many more in the flanking non-coding DNA.

2.2 Transcripts and HGVS names

A variant is named precisely with HGVS nomenclature: c. for the position in the coding DNA of a transcript (c.1 is the A of the ATG start codon), p. for the protein. For example c.681G>A and p.Pro227= (= means the amino acid is unchanged). Because positions depend on the transcript, an HGVS name always carries the transcript it refers to (e.g. ENST00000371321.3:c.681G>A, or the RefSeq NM_000769.4:c.681G>A). Genes on the minus strand are read in reverse complement: a genomic T>C can be an A>G in the gene.

Activity 2.2 · Name the protein-changing variants

Click the missense variant in CYP2C9 to open its details.
Now the two missense variants in CYP2C8. For each, open All transcripts in the details card.

What is the protein change of the CYP2C9 variant (HGVS p. name)?

p.Arg144Cys (R144C; c.430C>T, rs1799853). This is the defining variant of the CYP2C9*2 allele.

What are the two CYP2C8 protein changes? Why does the genomic change at chr10:96,798,749 read T>C when the codon changes A>G?

p.Arg139Lys (rs11572080) and p.Lys399Arg (rs10509681) – together they define CYP2C8*3. CYP2C8 lies on the minus strand, so the gene's sequence is the reverse complement of the reference: T>C on the plus strand is A>G in the codon (AAA → AGA, Lys → Arg).

The original version of this practical said the CYP2C8 variant is at “position 349 or 399” because there are two splice variants. What do the transcripts in VEP's details show, and why does it matter which one you report?

In the canonical transcript (ENST00000371270, the 490-residue protein – RefSeq NM_000770) the change is p.Lys399Arg. In another protein-coding transcript, ENST00000535898, the same DNA change is p.Lys297Arg, and in several non-coding or nonsense-mediated-decay transcripts it has no protein effect at all. The number depends on the isoform: the RefSeq transcripts give Lys399 (NM_000770), Lys329 (NM_001198853 and NM_001198855) and Lys297 (NM_001198854). None gives 349 – the older handout’s “349” was simply a mistake, which shows why a residue number is meaningless without its transcript. Always give the transcript with an HGVS name, and use the canonical (MANE Select) transcript unless there is a reason not to; PharmVar, ClinVar and the literature use the reference isoform.

How common is it?

≈ 15 minWorkbench: Variants

The single most useful filter for rare disease is population frequency. gnomAD (the Genome Aggregation Database) collects exomes and genomes from people without severe childhood disease. The frequencies in this table come from gnomAD version 4.1 – 730,947 exomes and 76,215 genomes – which Ensembl maps onto GRCh37 for its VEP service. The gnomAD website shows version 4 on GRCh38; to look up a GRCh37 position there directly, use the older version 2.1 (125,748 exomes and 15,708 genomes), so its numbers will differ a little from these.

Activity 3.1 · Sort by frequency

Click the gnomAD AF heading to sort by allele frequency, lowest first. AF is the fraction of all chromosomes in gnomAD that carry the ALT allele (a dash means it is not in gnomAD).
Set Population frequency to < 1%.
Reset the filter (Show all), then open CYP2C19*2 (chr10:96,541,616) and look at its frequency by ancestry.

How many of NA12878's variants have an allele frequency below 1%?

6 of 219 – and none changes a protein. Three are rare single-base changes in introns (two in MYOF, one in BLNK); two are the novel insertions/deletions in repeats, which gnomAD has never seen; and one, a single-base insertion in a non-coding transcript of ZNF518A, has no gnomAD frequency even though the 1000 Genomes Project finds it in everyone (open its details) – a reminder that “no frequency in gnomAD” does not always mean rare. Every protein-changing variant NA12878 has in this region is common – which is what a healthy genome looks like.

In which ancestry group is CYP2C19*2 most common in gnomAD exomes, and how does that compare with non-Finnish Europeans?

In gnomAD exomes the A allele is most frequent in South Asians (32%) and East Asians (30%) – about twice its frequency in non-Finnish Europeans (15%); African 18%, Latino/admixed American 10%, all 16%. Because everyone has two copies, 1 − (1 − q)² of people carry at least one: about half of East and South Asian people, and about a quarter of Europeans.

A variant is found in 8% of healthy people in gnomAD. For a rare, severe childhood disease this variant is…

a strong candidate, because it changes the protein almost certainly not the cause impossible to judge without a structure

So why can a common variant such as CYP2C19*2 matter so much for drug response?

Drug response is not a disease that natural selection has acted on: before modern medicine, being slow to activate clopidogrel cost nothing, so no-function alleles of drug-metabolising enzymes could drift to high frequency. The effect only appears when a person takes the drug. Pharmacogenomics is therefore mostly about common variants with large effects on a specific drug, while severe Mendelian disease is mostly caused by rare variants.

Predicting damage

≈ 15 minWorkbench: Variants

For most variants nobody has measured the effect, so computer predictions are used as (weak) evidence.

Tool	What it uses	How to read it
SIFT	How conserved the amino acid is across related proteins	Score 0–1; below 0.05 = deleterious, 0.05 or above = tolerated
PolyPhen-2	Conservation plus structural features of the position	Score 0–1; benign, possibly damaging, probably damaging (> 0.9)
CADD	Combines many annotations, for every kind of variant	PHRED-scaled rank: 10 = top 10%, 20 = top 1%, 30 = top 0.1% most deleterious of all possible changes
SpliceAI	A deep neural network trained on the sequence around splice sites	Delta score 0–1 for gain or loss of an acceptor or donor; ≥ 0.5 = likely to change splicing

Activity 4.1 · Compare the predictions

Show only HIGH and MODERATE impact variants again and compare the SIFT, PolyPhen and CADD columns for the three missense variants in CYP2C9 and CYP2C8. Hover over a prediction to see its score.
Sort all variants by CADD (highest first). Which variants score highest? Sort by SpliceAI too.

Do SIFT and PolyPhen-2 agree about CYP2C9 p.Arg144Cys? What does that tell you about using prediction tools?

No. SIFT: tolerated (0.05 – right on the threshold); PolyPhen-2: probably damaging (0.986); CADD 21.2 (top 1%). CYP2C9*2 is in fact a well-established decreased-function allele (CPIC), so here SIFT is wrong. Predictors often disagree, and each is wrong for a sizeable fraction of variants: SIFT relies on conservation alone, PolyPhen-2 was trained on known disease and neutral variants, and CADD learned to separate variants that survived human evolution from simulated new mutations – none of them measures drug metabolism. Use them as supporting evidence only (ACMG criteria PP3/BP4), never as proof – and prefer experimental evidence when it exists.

Which variant has the highest SpliceAI score, and what is its VEP consequence?

CYP2C19 chr10:96,541,616 G>A (rs4244285, *2): SpliceAI 0.99. Yet VEP calls it a synonymous variant (LOW impact) and its CADD score is only 14.6. The next highest SpliceAI score is 0.21. Chapter 5 explains.

A “silent” variant that is not silent

≈ 15 minWorkbench: IGV, Variants

VEP calls CYP2C19*2 (c.681G>A) a synonymous variant with LOW impact: CCG and CCA both encode proline 227. Yet people with two copies make essentially no CYP2C19 enzyme. The answer lies in splicing.

Activity 5.1 · Look at the exon boundary

In IGV, show the start of exon 5 of CYP2C19, with the variant: Zoom in until you can read the bases.
Find the end of intron 4 (the thin line) and the start of exon 5 (the thick box). The last two bases of the intron are AG – the splice acceptor. Now look 38 bases into the exon, at the variant.

How CYP2C19*2 works. Genomic sequence (plus strand, hg19 chr10:96,541,578–96,541,635). Based on de Morais et al. (1994) J Biol Chem 269:15419–15422; sequences from hg19 and RefSeq NM_000769.

Activity 5.2 · What the predictors say

Open the details of chr10:96,541,616 in the Variants tab and read the SpliceAI line.

What are SpliceAI's acceptor-gain and acceptor-loss scores, and at which positions? How do they match the figure?

Acceptor gain 0.99 at +2 bp and acceptor loss 0.89 at −38 bp (relative to the variant). SpliceAI predicts exactly what the figure shows: the normal acceptor at the start of exon 5 (38 bases before the variant) is lost, and a new one appears 2 bases after the variant, just after the new AG. Exon 5 then starts at c.683 instead of c.643: 40 bases shorter.

A lab filters its variants to keep only HIGH and MODERATE impact. What happens to CYP2C19*2, and how could the filter be improved?

It is thrown away – VEP classes it as synonymous (LOW), and its CADD score (14.6) is unremarkable. Better filters keep variants with a high splice-prediction score (e.g. SpliceAI ≥ 0.5), keep “splice region” variants for review, and always check known pathogenic or pharmacogenetic variants (ClinVar, PharmGKB/ClinPGx) whatever their predicted consequence. Chapter 6 lets you try both filters.

NA12878 is heterozygous. Roughly how much working CYP2C19 would you expect her to make, compared with someone with two normal copies?

None About half The same

Prioritising variants

≈ 20 minWorkbench: Variants

A diagnostic exome has 20,000–50,000 variants; a lab must end with a handful to look at by eye. Each filter answers a question about the case. The funnel above the table shows how many variants survive each step.

Activity 6.1 · Three funnels

Show the classic rare-disease funnel: keep variants that change the protein (HIGH or MODERATE impact), then only rare ones. Write down the number at each step.
Now keep predicted splice changes too.
Finally, ask the pharmacogenomic question: known variants that change the protein or its splicing in the CYP2C genes.

How many variants survive the classic rare-disease funnel? Is that what you would expect for NA12878 in this region? What stage would a real diagnostic funnel add next?

None. 12 variants change a protein (all missense), adding predicted splice changes brings in CYP2C19*2 (13), but every one is common (allele frequency ≥ 1%; most above 10%). That is what we expect: NA12878 is healthy, and this is 3 Mb, about 0.1% of her genome. A real diagnostic funnel would next ask whether a gene fits the patient's condition (a gene panel, or phenotype terms such as HPO), whether the zygosity fits the inheritance (one variant for a dominant or new disorder, two for a recessive one), and whether ClinVar already knows the variant. You will build exactly this for the coursework patient.

Which variants remain in the pharmacogenomic funnel? Give the gene and protein change for each.

Four, all heterozygous: CYP2C19 c.681G>A p.(Pro227=) (rs4244285 – *2, splicing defect); CYP2C9 p.Arg144Cys (rs1799853 – *2); CYP2C8 p.Arg139Lys (rs11572080) and p.Lys399Arg (rs10509681) – together CYP2C8*3.

6.2 From variants to star alleles and phenotypes

Pharmacogenes are described with star alleles – named haplotypes curated by PharmVar. A person's two star alleles (the diplotype, e.g. *1/*2) are translated into a predicted phenotype (normal, intermediate or poor metaboliser), which guidelines from CPIC and the Dutch DPWG turn into prescribing advice. *1 is the reference allele: it is assigned when none of the defining variants is found.

Gene	Variant(s) found	Allele	Function (CPIC)
CYP2C19	rs4244285 c.681G>A (splicing defect)	*2	no function
CYP2C9	rs1799853 p.Arg144Cys	*2	decreased function
CYP2C8	rs11572080 p.Arg139Lys + rs10509681 p.Lys399Arg	*3	reduced activity for some substrates in the laboratory; no CPIC dosing guideline

CYP2C8*3 needs both changes on the same chromosome. NA12878 is heterozygous (0/1) for each. Can your data tell you whether they are on the same copy of chromosome 10? What could?

No. The genotypes are unphased (written with /): 0/1 and 0/1 fit both “both on one chromosome” (*1/*3) and “one on each” (two different rarer alleles). The variants are about 28 kb apart, far beyond a 76-base read or a ~170-bp fragment, so no read covers both. Options: population linkage (in Europeans these two almost always travel together, so *1/*3 is by far the most likely), family data (NA12878's parents have been sequenced – trio phasing), or long-read or linked-read sequencing that reads whole haplotypes.

NA12878 is CYP2C19 *1/*2. According to CPIC (2022), what does this mean if she needed antiplatelet therapy after a stent (PCI) for an acute coronary syndrome?

Standard-dose clopidogrel, as for everyone else She is an intermediate metaboliser: prefer prasugrel or ticagrelor if there is no contraindication Double the dose of clopidogrel

Extension: what the exome cannot see Extension

The increased-function allele CYP2C19*17 is defined by a promoter variant, rs12248560 (chr10:96,521,657 in hg19), about 800 bases before the start codon. In IGV, go to . Are there reads? What does that mean for NA12878's diplotype – and for any gene test based on exome data?

Can you rule out *17 from these data?

No. No reads cover chr10:96,521,657 – the promoter lies outside the exome capture targets. So NA12878 could be *1/*2 or *17/*2 (CPIC also classes *2/*17 as an intermediate metaboliser, so the clopidogrel advice would not change here). This is why clinical pharmacogenetic tests genotype the defining positions directly. “No variant called” only means “reference” where there are reads: absence of evidence is not evidence of absence.

The variants in 3D

≈ 20 minWorkbench: 3D viewer

A structure turns a residue number into a place: buried or on the surface, next to the haem or far away. All three CYP2C enzymes have crystal structures. (The viewer works like the one in the molecular graphics practical: left-drag rotates, scroll zooms, right-drag moves; click a residue to select it; the command line is at the bottom.)

7.1 CYP2C19 and the *2 truncation

Activity 7.1 · 4GQS

Load CYP2C19 (PDB 4GQS: 2.87 Å, four copies of the enzyme with a benzbromarone-like inhibitor).
Keep one copy and show the haem:
Colour what a *2 protein would keep: residues up to 214 (blue), the 20 frameshifted residues 215–234 (orange – their real sequence would be different), and the part that is lost, 235–490 (grey):
Rotate the structure. Where is the haem? Where is Cys435 (yellow), and what does it do?

Using the picture, explain why the *2 protein cannot work even if it folded.

The haem sits in the core of the second half of the protein, held from below by the thiolate of Cys435, which binds the iron (Cys435 Sγ–Fe 2.3 Å in 4GQS) – the defining ligand of all P450s, in the conserved FxxGxRxCxG haem-binding loop. The *2 protein stops at residue 234: it lacks Cys435, the I-helix above the haem and most of the haem pocket (grey), and its last 20 residues are wrong (orange). It cannot bind haem, so it has no catalytic activity – and a protein missing half its fold is likely to be degraded anyway.

Extension: CYP2C19*8 Extension

UniProt lists CYP2C19*8 (p.Trp120Arg) as an inactive allele. How close is it to the haem, and why might Arg be tolerated badly here?

What does Trp120 touch?

Its side chain lies against the haem (closest atoms about 2.7 Å – a hydrogen bond to a haem propionate). Replacing the flat, hydrophobic indole ring with a long, positively charged arginine would disturb how the haem is held, destabilising haem binding and the active site.

7.2 CYP2C9 Arg144

Activity 7.2 · 1R9O

Load CYP2C9 with the NSAID flurbiprofen bound (PDB 1R9O, 2.0 Å) and show Arg144:
Measure from the Arg144 Cα to the haem iron:

How far is the Arg144 Cα from the haem iron (Å)? Is Arg144 in the active site?

About 18 Å, and its side chain points out into the solvent. Arg144 is on the surface, on the side of the enzyme below the haem (the “proximal” face) – far from where flurbiprofen binds. That face is thought to be where cytochrome P450 reductase docks to deliver electrons, which would explain how R144C reduces activity without touching the substrate pocket. Structure explains why SIFT (conservation only) and PolyPhen-2 disagreed: the position is not deeply conserved, but it matters for a protein partner.

7.3 CYP2C8 Arg139 and Lys399

Activity 7.3 · 2NNI

Load CYP2C8 with montelukast (PDB 2NNI) and show both residues of *3:

Where are Arg139 and Lys399 relative to the haem and the drug? Are the two changes conservative?

Both are on the surface: their closest atoms are about 14 Å (Arg139) and 21 Å (Lys399) from the haem (Cα to iron: 22 and 27 Å), and further still from montelukast – neither lines the active site. Both are conservative swaps between the two positively charged residues (Arg↔Lys). Consistent with this, CYP2C8*3 changes activity modestly and in a substrate-dependent way (reduced for paclitaxel in the laboratory), rather than abolishing it.

Reporting variants

≈ 10 min

A variant in a report must be unambiguous and checkable. Your coursework asks for a results table and figures of exactly this kind.

Chr	Position (hg19)	dbSNP	Ref/Alt	Genotype (reads ref/alt)	Gene	HGVS	Consequence	gnomAD AF	Predictions
10	96,541,616	rs4244285	G/A	het (44/48)	CYP2C19	NM_000769.4:c.681G>A, p.(Pro227=)	synonymous; aberrant splicing (*2)	16% (gnomAD exomes)	SpliceAI 0.99 (acceptor gain)

Genome build in the heading or legend (hg19/GRCh37) – positions mean nothing without it.
Transcript with the HGVS name; zygosity and the read counts that support it.
Evidence: frequency, predictions, database entries (dbSNP, ClinVar, OMIM, PharmGKB), literature.
Figures with legends that say what, where and how: “IGV view of chr10:96,541,596–96,541,636 (hg19) showing NA12878 exome reads; 58 of 109 reads carry A at c.681 of CYP2C19 …”.

Activity 8.1 · Take your results with you

Press Table above the variant table to download the variants currently shown as a tab-separated file (it opens in Excel).

Check your understanding

You annotate hg19 variants with the GRCh38 version of VEP. What happens?

VEP refuses and asks for GRCh37 You get annotations for the wrong places, usually without an error Nothing important – only the numbers differ

Which consequence has HIGH impact?

missense_variant stop_gained splice_region_variant synonymous_variant

A missense variant has CADD 28, SIFT “deleterious”, PolyPhen “probably damaging”. What can you conclude?

It is pathogenic The predictions support – but do not prove – a damaging effect Nothing – scores are meaningless

A rare heterozygous stop-gained variant is found in a child with a dominant-looking disorder. Which extra information would help most?

The parents' genotypes and whether the gene is known to cause disease when one copy is lost A higher QUAL score A list of the child's common variants

Now: your coursework

You have run every step your coursework needs: align (done for you), call, filter, annotate, prioritise, inspect and visualise. The coursework workbench runs the real bcftools and VEP on the patient's exome – on your own computer, without uploading the reads.

Reference

Databases and tools

Resource	Use it for	Link (GRCh37 where it matters)
Ensembl VEP	Annotating variants	grch37.ensembl.org → VEP
gnomAD	Population frequencies	gnomad.broadinstitute.org (dataset v2.1.1 for GRCh37 positions; v4 uses GRCh38)
dbSNP	rs identifiers, positions in both builds	ncbi.nlm.nih.gov/snp
ClinVar	Clinical interpretations submitted by labs	ncbi.nlm.nih.gov/clinvar
OMIM	Genes and Mendelian disorders	omim.org
ClinPGx (PharmGKB) · CPIC · PharmVar	Pharmacogenomic evidence, guidelines, star alleles	clinpgx.org · cpicpgx.org · pharmvar.org
UniProt · PDB · AlphaFold DB	Protein function, structures, models	uniprot.org · rcsb.org · alphafold.ebi.ac.uk

Credits and references

Annotation: Ensembl VEP REST service, GRCh37 (McLaren W et al. (2016) Genome Biol 17:122), with gnomAD v4.1 exome and genome frequencies mapped to GRCh37 by Ensembl (gnomAD: Karczewski KJ et al. (2020) Nature 581:434–443), SIFT (Ng PC & Henikoff S (2003) Nucleic Acids Res 31:3812–3814), PolyPhen-2 (Adzhubei IA et al. (2010) Nat Methods 7:248–249), CADD (Rentzsch P et al. (2019) Nucleic Acids Res 47:D886–D894) and SpliceAI (Jaganathan K et al. (2019) Cell 176:535–548).
CYP2C19*2: de Morais SM et al. (1994) J Biol Chem 269:15419–15422. Clopidogrel guideline: Lee CR et al. (2022) Clin Pharmacol Ther 112:959–967. Variant interpretation: Richards S et al. (2015) Genet Med 17:405–424.
Structures: 4GQS – Reynald JA et al. (2012) J Biol Chem 287:44581–44591; 1R9O – Wester MR et al. (2004) J Biol Chem 279:35630–35637; 2NNI – Schoch GA et al. (2008) J Biol Chem 283:17227–17237. Molecular graphics: NGL Viewer (Rose AS et al. (2018) Bioinformatics 34:3755–3758; MIT licence).
Reads: 1000 Genomes Project NA12878 exome (SRR098401). Genome browser: igv.js (MIT licence).
Adapted from the module's cytochrome P450 IGV tutorial, VEP walkthrough and molecular graphics tutorial.