From $19, paid once and yours to keep. See what your file reads first. Check your file →
DeutschFrançaisEspañolPortuguês Sample report Support Log in
aimosti Check your file

Learn Guide

CYP2D6 from a whole genome you already have: which file holds the answer, and how to get it read

If you bought a 30x genome, you were handed a VCF, and a BAM or CRAM either came with it or can be requested. The VCF is the small file and the easy one to upload. For CYP2D6, the gene behind codeine, tramadol, tamoxifen and several antidepressants, the answer is in the other file, and on one of our own test genomes the VCF alone missed a whole copy of the gene without saying so.

Key takeaways

  • PharmCAT's documentation does not recommend calling CYP2D6 from a VCF, and says an ultrarapid metabolizer cannot be called from the SNPs and indels a VCF holds[1].
  • On one of our own 30x test genomes, PyPGx run on the VCF alone returned *1/*41. Run on the same genome's aligned reads, it returned *1x2/*41, a whole-gene duplication, and the VCF run gave no warning that a copy was missing.
  • Both results fall in CPIC's normal band, at activity scores of 1.25 and 2.25. The same missed copy on a *1/*1 genome would report an ultrarapid metabolizer (3.0) as normal (2.0), and CPIC advises ultrarapid metabolizers to avoid codeine and tramadol[4, 6].
  • Published callers read CYP2D6 from a 30x BAM or CRAM, among them Cyrius, Aldy, Stargazer and PyPGx. In an independent test on 75 reference genomes the best of three matched 67 diplotypes, so a reads-based call is better evidence and still not a certainty[9].
  • Our Deep Read calls CYP2D6 from the reads. The browser sends only the parts of the file the panel needs, and the card gives no phenotype when the gene was poorly covered or an allele is outside its activity table.

CYP2D6 varies in two ways. Single changes alter what one copy of the gene does, and whole copies come and go: a person can carry a deletion, a duplication, or a hybrid with the neighbouring CYP2D7 gene. A VCF records the letters at positions where a genome differs from the reference, so it holds the first kind of change and not the second. PharmCAT, a pharmacogenomics tool that turns genotypes into prescribing recommendations, says so in its documentation: it does not recommend calling CYP2D6 from a VCF, and an ultrarapid metabolizer cannot be called from the SNPs and indels a VCF holds[1]. The aligned reads in a BAM or CRAM keep the depth of coverage, and depth is how a caller counts copies.

What each file from a 30x genome can say about CYP2D6

A sequenced genome usually arrives as several files, each the same genome at a different stage of processing. The FASTQ holds the raw reads. The BAM, or its compressed form the CRAM, holds those reads placed on the reference genome, and for a 30x genome it runs to tens of gigabytes. The VCF lists the positions where you differ from the reference and is usually under a gigabyte. Our guide to VCF, gVCF, BAM and CRAM files goes through what each one keeps.

For most pharmacogenes the variation that matters is a short list of single changes, and a VCF can carry those. CYP2D6 is the exception. Whole copies of the gene are deleted, duplicated or fused with CYP2D7, and the number of working copies moves the result as far as any single change does. A VCF has no field for that number. PharmCAT's documentation puts it bluntly: "we do NOT recommend calling CYP2D6 from VCF", because the effect of structural and copy-number variation on the phenotype "is beyond the scope of what can be called from SNPs or INDELs in a VCF file"[1]. Our star alleles guide shows three people whose CYP2D6 looks the same at every position and differs only in the number of copies.

Table 1. What each file from a 30x genome can tell you about CYP2D6
FileWhat it holdsCYP2D6
FASTQRaw reads, not yet placed on the genomeUsable only after alignment; we do not read it
BAM or CRAMThe reads aligned to the reference genomeSingle changes and the copy count, from read depth
VCFPositions where you differ from the referenceSingle changes only; a caller has to assume two copies
gVCFThe same, plus blocks showing where the reference was readSingle changes only; still no copy count

Source: The files a 30x provider usually delivers. What each format keeps is set out in our guide to the four formats.

What the VCF missed on one of our test genomes

We tested this on one of our own 30x genomes, for which we hold both the VCF and the aligned reads. PyPGx, the caller our Deep Read uses, was run on each file. On the reads it compares the depth over CYP2D6 with the depth over a control gene and classifies the copy-number profile with a trained model[2]. Given the VCF alone it has no depth to compare, so it reads the letters and assumes the usual two copies.

Table 2. CYP2D6 on one of our 30x test genomes, called by PyPGx from each file
InputDiplotypeStructural callActivity scoreCPIC group
VCF alone*1/*41None reported1.25Normal metabolizer
Aligned reads*1x2/*41WholeDup12.25Normal metabolizer

Source: PyPGx 0.26.0 on the VCF and on the reads of the same genome, September 2026. The scores and groups are the ones our report computes from CPIC's activity values.

WholeDup1 is PyPGx's name for a whole extra copy of the gene on one chromosome[3], three copies of CYP2D6 in all, and the x2 in *1x2 records it. Both answers fall in the normal band, which the CPIC and Dutch Pharmacogenetics Working Group consensus sets at activity scores from 1.25 to 2.25[4]. CPIC gives *41 a value of 0.25 and a duplicated *1 a value of 2[5]. The missed copy moved the score from the bottom of the normal band to its top, and the VCF run said nothing about it: no copy-number column, no warning. Its output looked as finished as the correct one.

On this genome the error did not cross a line between groups. It takes very little to cross one. The table applies the same arithmetic to three other genotypes, using the activity values our report adds up.

Table 3. The same kind of miss on three other CYP2D6 genotypes
True diplotypeScoreCPIC groupWithout the copy countScoreCPIC group
*1/*1x23.0Ultrarapid metabolizer*1/*12.0Normal metabolizer
*2x2/*42.0Normal metabolizer*2/*41.0Intermediate metabolizer
*1/*51.0Intermediate metabolizer*1/*12.0Normal metabolizer

Source: Activity values from CPIC's CYP2D6 table as our report applies them, grouped by the CPIC and DPWG consensus[4, 5]. The second row is PharmCAT's own example[1]. In the third, *5 is a deleted copy, and a VCF shows the remaining copy's letters as if there were two of it.

The first row matters most for a prescription. For an ultrarapid metabolizer, CPIC's opioid guideline says to "avoid codeine use because of potential for serious toxicity" and to avoid tramadol for the same reason, because the enzyme turns both into their stronger forms faster than expected[6]. A call made from a VCF cannot put anyone in that group, which is the limit PharmCAT describes[1].

Before you upload: check the file you have

Whether you already hold the aligned reads depends on the provider. Nebula's customer terms, quoted on our Nebula page, make the VCF downloadable from the portal and release the CRAM on request, for a data-transfer fee. Dante Labs' product pages promise access to the BAM, the VCF and the FASTQ from its Genome Manager portal, as our Dante Labs page quotes. Sequencing.com's product pages describe a BAM aligned to GRCh38 alongside the VCF; see our Sequencing.com page.

Four things to check before CYP2D6 can be read

BAM or CRAM
Both hold the same aligned reads. A CRAM is smaller because it stores each read as its differences from one exact reference assembly, so it can be decoded only with that same assembly. Our file check reads the header in your browser, names the assembly a CRAM needs and says whether we hold it.
The index
A BAM comes with a .bai index and a CRAM with a .crai. The index lets software go straight to CYP2D6 on chromosome 22 without reading the whole file, so ask for it with the file.
GRCh37 or GRCh38
Older genomes are often aligned to GRCh37 and newer ones to GRCh38, and the build is written in the file's header. PyPGx works with both builds[2], and our Deep Read reads GRCh37 and GRCh38 files.
Depth
Copy number is read from depth, so the reads have to be deep enough to count. A 30x genome is. A low-pass genome sequenced 1x or 2x is not, and PharmCAT's documentation strongly discourages calling CYP2D6 from low-coverage genomes or exomes[1]. Deep Read refuses a file whose mean depth is below 5 reads instead of returning a guess.

You do not have to send us the whole file. The browser reads the index, works out which parts of the file the Deep Read panel needs and uploads only those, a few percent of a 30x genome. The reads are processed in Finland and deleted when the run finishes, whether it succeeded or failed.

Running a CYP2D6 caller yourself

If you can work at a command line, you can call CYP2D6 from your own reads with published research software. Each tool in the table reads a BAM or CRAM from a whole genome and estimates copy number from depth. They differ in method and in how they were tested. All of them need the reference genome your file was aligned to, and each has its own installation steps and licence terms.

Table 4. Published callers that read CYP2D6 from aligned whole-genome data
ToolHow it reads CYP2D6What has been measured
CyriusWritten for CYP2D6 in whole-genome data, structural variants included96.5% concordance with truth genotypes in its paper, against 84 to 86.8% for the methods it was compared with[7]
AldyWorks out the number of copies of a gene and the sequence of each copy, as an optimisation problem[8]66 of 75 GeT-RM reference samples matched in an independent comparison[9]
StargazerCopy number for CYP2D6 and CYP2D7 separately, from read depth, plus single changes and indels[10]99.0% concordant on 32 HapMap trios in its paper[10]; 67 of 75 GeT-RM samples matched[9]
PyPGxDepth against a control gene, classified by a trained model; 88 pharmacogenes, GRCh37 or GRCh38[2]Written by Stargazer's author; the documentation reports near-complete agreement with GeT-RM samples on a targeted sequencing panel[2]

Source: Each tool's paper or documentation. The GeT-RM counts come from one independent comparison of three callers on the same 75 genomes[9].

These tools return a diplotype, such as *1x2/*41. Turning it into a recommendation takes one more step: look the diplotype up in CPIC's tables, or hand it to PharmCAT as an outside call: its documentation describes that route, with another tool calling CYP2D6 and PharmCAT supplying the phenotype and the recommendations[1]. PharmCAT's own CYP2D6 page uses StellarPGx, another reads-based caller, for its examples[1].

A call from the reads is better evidence than one from a VCF, and it is still not a certainty. The independent comparison ran three callers on 75 whole genomes from GeT-RM, a set of reference samples whose CYP2D6 genotypes laboratories have characterised. Stargazer matched 67 diplotypes, Aldy 66 and Astrolabe 54, and Astrolabe had the most trouble with hybrid rearrangements[9]. Where a result would change a prescription, the prescriber can order a clinical test that measures the gene directly.

What our Deep Read reports for CYP2D6

Deep Read is our panel read from aligned reads, and CYP2D6 is one of its 29 genes. It runs PyPGx on your BAM or CRAM and restates the result on one card per gene. The CYP2D6 guide goes through each step the run takes; the table lists what the card shows.

Table 5. The CYP2D6 card in Deep Read
What the card showsWhere it comes from
The diplotype, such as *1x2/*41PyPGx run on your reads
The structural call, such as WholeDup1PyPGx's copy-number model, with depth compared against the control gene VDR
The activity score and the CPIC groupCPIC's activity values, added up and grouped by the CPIC and DPWG consensus[4, 5]
Guidance for codeine and tramadol, tamoxifen, fluvoxamine, paroxetine, venlafaxine and vortioxetineCPIC's guidelines, quoted for the group the card names
The share of the gene read by at least 10 readsMeasured on your reads; under 30 percent, the card gives no phenotype
Indeterminate, with no phenotypeAn allele outside our activity table, such as the hybrids *13, *36 and *68

Source: The report's CYP2D6 content and engine, pinned by this article's tests.

Deep Read adds the reads-only part to our whole-genome report, which is read from your VCF: clinical findings, carrier status, ancestry and the pharmacogenes a VCF can carry. On its own it is added to a report you already have, or it can be bought together with the whole-genome report; pricing lists both. The file check reads either file in your browser and says what it supports before you pay.

What Aimosti would (and wouldn't) show you

Deep Read calls CYP2D6 from your BAM or CRAM with PyPGx. The card shows the diplotype, the structural call in PyPGx's own name for it (WholeDup1 for a duplicated copy), the activity score and the CPIC group, with CPIC's guidance for codeine and tramadol, tamoxifen, fluvoxamine, paroxetine, venlafaxine and vortioxetine. It gives no phenotype when less than 30 percent of the gene is covered by at least 10 reads, and none when an allele is outside its activity table, as the hybrids *13, *36 and *68 are. A file whose mean depth is below 5 reads is refused before any calling. CYP2D6 is never called from your VCF.

What we won't claim

We won't call CYP2D6 from a VCF or a gVCF, put a phenotype on an allele our table cannot score, or tell anyone to start, stop or change a medicine. CPIC's guidelines are written for prescribers, and a result that would change a prescription should be confirmed by a clinical test.

Bottom line. For CYP2D6 the file to bring is the BAM or CRAM. A VCF can name the letters in each copy of the gene but not how many copies there are, and on one of our own test genomes that cost a whole working copy with no warning. Read from the aligned reads, by our Deep Read or by a caller you run yourself, the gene comes back with a copy count, and a prescriber can look the diplotype up in CPIC's tables.

Questions people ask

Can I get CYP2D6 from my Nebula VCF?

Software will produce a diplotype from it, but not a copy count, and the count is part of the answer. PharmCAT's documentation does not recommend calling CYP2D6 from a VCF[1]. Nebula's terms, as our Nebula page quotes them, release the CRAM on request for a data-transfer fee, and the CRAM is the file that settles CYP2D6.

Do I need the BAM and the VCF, or just one of them?

For CYP2D6, the BAM or CRAM: our Deep Read calls the gene from the reads only. The rest of our whole-genome report is read from the VCF. Deep Read is added to a report you already have, or bought together with the whole-genome report, and then you upload both files.

Is a CRAM as good as a BAM for CYP2D6?

Yes. It holds the same aligned reads in a smaller file. The one catch is that a CRAM can be decoded only with the exact reference assembly it was written against, so the file check names the assembly before you pay.

Can a low-pass genome (1x or 2x) give a CYP2D6 result?

Not a trustworthy one. Copy number is read from depth, and PharmCAT's documentation strongly discourages calling CYP2D6 from low-coverage whole genomes[1]. Deep Read refuses a file whose mean depth is below 5 reads.

What does WholeDup1 mean on a CYP2D6 result?

It is PyPGx's name for a whole extra copy of the gene on one chromosome, which makes three copies in all[3]. In the diplotype it shows up as x2 after the duplicated allele, as in *1x2/*41.

Should I change a medicine because of my CYP2D6 result?

Not on your own. CPIC's guidelines are written for prescribers, who weigh the result with everything else they know about you. If a result would change a prescription, the prescriber can confirm it first with a clinical test.

References

  1. Calling CYP2D6. PharmCAT documentation, 2026. Read 12 November 2026. "While PharmCAT supports CYP2D6, we do NOT recommend calling CYP2D6 from VCF"; CYP2D6 ultrarapid metabolizers "cannot be called using only SNPs and INDELs in a VCF file"; *2x2/*4 misread as *2/*4; the outside-call route; StellarPGx used for the examples; "We strongly discourage using whole-exome sequencing or low coverage whole-genome sequencing data to call CYP2D6".
  2. Lee SB. PyPGx documentation: README. PyPGx, 2026. Read 12 November 2026. Structural variation detected for NGS data with a machine-learning approach; a control gene's read depth converts target depth to copy number; compatible with GRCh37 and GRCh38; 88 pharmacogenes; development heavily inspired by Stargazer; almost 100% concordance with GeT-RM genotypes in the ClinPharmSeq targeted-panel study.
  3. Lee SB. PyPGx documentation: Genes (SV summary for CYP2D6). PyPGx, 2026. Read 12 November 2026. WholeDup1 is shown with the duplicated allele *4x2 in *2/*4x2; WholeMultip1 with *1x3.
  4. Caudle KE, Sangkuhl K, Whirl-Carrillo M, et al. Standardizing CYP2D6 genotype to phenotype translation: consensus recommendations from the Clinical Pharmacogenetics Implementation Consortium and Dutch Pharmacogenetics Working Group. Clinical and Translational Science, 2020. doi:10.1111/cts.12692 Table 2, consensus: poor 0; intermediate 0 < x < 1.25; normal 1.25 to 2.25; ultrarapid above 2.25.
  5. CYP2D6 allele functionality table. Clinical Pharmacogenetics Implementation Consortium (CPIC), 2026. Read 12 November 2026. Activity values: *1 1.0; *1x2 2.0; *2x2 2.0; *41 0.25; *4, *5, *13, *36 and *68 0.0.
  6. Crews KR, Monte AA, Huddart R, et al. Clinical Pharmacogenetics Implementation Consortium guideline for CYP2D6, OPRM1, and COMT genotypes and select opioid therapy. Clinical Pharmacology & Therapeutics, 2021. doi:10.1002/cpt.2149 Ultrarapid metabolizer: "Avoid codeine use because of potential for serious toxicity"; "Avoid tramadol use because of potential for toxicity". Both strong recommendations.
  7. Chen X, Shen F, Gonzaludo N, et al. Cyrius: accurate CYP2D6 genotyping using whole-genome sequencing data. The Pharmacogenomics Journal, 2021. doi:10.1038/s41397-020-00205-5 Abstract: 96.5% concordance with truth genotypes, against 84 to 86.8% for existing methods; the authors report 99.3% after later improvements.
  8. Numanagić I, Malikić S, Ford M, et al. Allelic decomposition and exact genotyping of highly polymorphic and structurally variant genes. Nature Communications, 2018. doi:10.1038/s41467-018-03273-1
  9. Twesigomwe D, Wright GEB, Drögemöller BI, et al. A systematic comparison of pharmacogene star allele calling bioinformatics algorithms: a focus on CYP2D6 genotyping. npj Genomic Medicine, 2020. doi:10.1038/s41525-020-0135-2 75 GeT-RM whole genomes: Stargazer 67, Aldy 66, Astrolabe 54 diplotypes concordant; Astrolabe had limitations especially in calling hybrid rearrangements.
  10. Lee SB, Wheeler MM, Patterson K, et al. Stargazer: a software tool for calling star alleles from next-generation sequencing data using CYP2D6 as a model. Genetics in Medicine, 2019. doi:10.1038/s41436-018-0054-0 Abstract: paralog-specific copy number from read depth; 99.0% concordant with orthogonal methods on 32 HapMap trios; 28.1% of samples had structural variation.

Last reviewed . Every number on this page links to the source it comes from; if one of them has moved, tell us.

This one is read from your aligned reads. See what Deep Read produces.

See a sample report See Deep Read · $49