Genetic Testing for ADHD Medications: What It Can and Cannot Tell You

Woolf Software
pharmacogenomics adhd genomics cyp2d6 sequencing
A feathered reptilian specimen on a black plinth in a dark studio, one amber-glowing organ visible through translucent skin on its flank.

If you are wondering whether a commercial pharmacogenomic panel can tell you which ADHD medication to take, the short answer is no. No genotype currently predicts whether methylphenidate or amphetamine will work for a given person. The panels that imply otherwise are extrapolating from antidepressant pharmacology. The evidence there is thinner than the marketing suggests, and most of the signal comes from a handful of drug-metabolizing enzymes that barely touch stimulants.

There is one genuine exception worth knowing about. Atomoxetine is a substrate of CYP2D6, a liver enzyme whose activity varies widely between people for genetic reasons. A person’s CYP2D6 metabolizer status changes atomoxetine plasma exposure by roughly an order of magnitude. That single gene-drug pair is the part of ADHD pharmacogenomics with real clinical guidance behind it. It also happens to be the part you can extract from your own whole-genome data if you know what you are doing.

The rest of this page explains why the efficacy claims fail, what the pharmacokinetic claims do support, and how to genotype the relevant loci yourself.

Why response prediction for stimulants does not work yet

It helps to start with why the research has not produced a usable predictor. The reasons are structural rather than a matter of waiting for one more study. ADHD medication response is a polygenic trait measured with considerable noise, and the studies attempting to map it have been small. The first genome-wide association study of methylphenidate response enrolled 187 children and found no variant reaching genome-wide significance. That is the expected outcome at that sample size for a trait with realistic effect sizes.1 Two decades of candidate-gene work on DRD4, DAT1/SLC6A3, ADRA2A, and COMT produced a literature of small positive findings that have not replicated consistently, and reviews of the field have been saying so since the mid-2000s.2 The broader clinical review of genetics in the ADHD clinic reaches the same conclusion. The heritability of ADHD itself is high, but that heritability has not translated into actionable predictors of drug response.3

Recent work has moved from single variants to machine learning applied to genome-wide data, which is the right direction. The results remain early. A 2025 deep-learning analysis of genome data identified new gene sets associated with pharmacological treatment response in ADHD and reported predictive performance above chance in held-out data.4 That is a research finding rather than a clinical test. None of it has been validated prospectively in an independent cohort, and none of the commercial panels available today implement anything resembling it.

The commercial panels also inherit a problem from psychiatry more broadly. Reviews of the evidence basis for pharmacogenetic testing in psychiatry find that the randomized trials supporting combinatorial panels are mostly in depression and often industry-sponsored. They typically show modest effects on symptom scales that shrink or disappear when the outcome is remission rather than a continuous score.5 Broader assessments of precision psychiatry make the same point in a different way. The pharmacokinetic gene-drug pairs stand on reasonably firm ground. The pharmacodynamic ones do not, including serotonin transporter promoter variants and HTR2A.67 ADHD stimulants sit almost entirely outside even the firm part of that ground.

The pharmacokinetics that are real

Having set aside efficacy prediction, it is worth being precise about what genotype does influence. That is how quickly the body clears each drug, and the picture differs from one medication to the next.

Methylphenidate is hydrolyzed primarily by carboxylesterase 1, encoded by CES1, rather than by the cytochrome P450 enzymes that dominate pharmacogenomic panels. The well-characterized CES1 variant G143E (rs71647871) reduces enzyme activity substantially but is rare. Its minor allele frequency is around 3 to 4 percent in European ancestry populations and lower elsewhere. Amphetamine is partly metabolized by CYP2D6 but is also excreted unchanged in urine to a degree that depends on urinary pH, which blunts the effect of genotype on exposure. Neither drug has a guideline from the Clinical Pharmacogenetics Implementation Consortium, the group that translates gene-drug evidence into dosing recommendations.

Atomoxetine does have such a guideline. It is cleared predominantly by CYP2D6, and poor metabolizers show area-under-the-curve exposures roughly ten times those of normal metabolizers, with correspondingly longer half-lives. CPIC publishes dosing guidance keyed to the CYP2D6 activity score, and the FDA label references metabolizer status. If you take one genotype away from this whole topic, it should be CYP2D6. Viloxazine, a newer nonstimulant, is a CYP2D6 substrate as well and is a strong CYP1A2 inhibitor. That inhibition matters more for drug interactions than for your own genotype.

Guanfacine is metabolized by CYP3A4. Common functional germline variation in that gene is limited in most populations, so genotype adds little beyond knowing which other medications you are taking.

One caveat applies to everything that follows. Any change to a prescribed medication, including a dose change based on metabolizer status, requires a prescribing clinician. Genotype informs that conversation rather than replacing it.

Calling these genes from your own data

If you have whole-genome sequencing, you can produce a more complete pharmacogenomic result than most commercial panels offer. Panels genotype a fixed list of variants, whereas sequencing sees everything in the interval. That includes rare loss-of-function alleles that no array covers. The cost of that completeness is that CYP2D6 is one of the hardest genes in the genome to call correctly.

The difficulty comes from the neighborhood. CYP2D6 sits next to two pseudogenes, CYP2D7 and CYP2D8, with which it recombines to form hybrid alleles. It also varies in copy number from zero to more than five, and duplications of a functional allele produce ultrarapid metabolizers. Short-read aligners mismap reads between CYP2D6 and CYP2D7 routinely, so a naive variant call at that locus will be wrong in a meaningful fraction of samples.

The tool we would use is Aldy, which calls star alleles from aligned reads. That includes copy number and hybrids. Here is a typical invocation on a 30x whole-genome BAM or CRAM aligned to GRCh38. The 30x figure means each position is covered by about thirty sequencing reads on average.

aldy genotype -p wgs -g CYP2D6 --genome hg38 \
  -o cyp2d6.aldy sample.cram

While you are in the data, run it for the other genes on the same panel as well: -g CYP2C19, -g CYP2C9, -g CYP3A4, -g CYP1A2. PharmCAT is the complementary tool. You feed it a VCF, the standard text format listing the variants found in a sample, restricted to the pharmacogenomic positions. It then produces a report mapping diplotypes to CPIC guidance. It also accepts an external CYP2D6 call from Aldy, so you are not relying on its own limited handling of that locus.

java -jar pharmcat-pipeline.jar -vcf sample.pgx.vcf.gz \
  -matcherHtml -reporterJson \
  -research cyp2d6

Two failure modes deserve attention before you trust the output. The first is coverage. At 30x mean genome coverage, the CYP2D6 region often runs below that average because of GC content and mapping quality filters, and Aldy will report low-confidence calls as a result. Check depth with samtools depth -r chr22:42126000-42131000 before trusting the diplotype. The second is structural variation. If Aldy reports a hybrid such as *68+*4, treat the activity score as provisional. Long-read sequencing resolves these cleanly. If CYP2D6 status bears on a decision you are making, a targeted long-read assay or a CLIA-validated clinical test is the right confirmation step. Research-grade calls from your own pipeline are not a clinical result and should not be handed to a prescriber as one.

Whole-genome data also lets you check the rare CES1 and SLC6A2 variants that no consumer panel includes. What you learn from them sits closer to a research observation than to a dosing input. Mutation burden analyses across psychiatric disorders using whole-genome sequencing illustrate both the value and the limits of this approach. Rare-variant signal exists, and it is not yet individually interpretable.8

What we would do

Our recommendation is to genotype CYP2D6 properly. Atomoxetine and viloxazine exposure depend on it, and the same call stays useful across many other drugs you may encounter over a lifetime. Genotype CYP2C19 and CYP2C9 while you are already in the data, since they matter for common comedications. We would skip the combinatorial “psychotropic” interpretation layer entirely. The proprietary algorithms combine pharmacodynamic markers of weak evidence into a red-yellow-green display that conveys more confidence than the underlying data supports.5 Bring the diplotypes, rather than the color-coded report, to the clinician who manages your medications.

For choosing among stimulants, careful titration under clinical supervision remains the method, and genotype does not shorten it.

Questions people also ask

Is a GeneSight test worth it? For ADHD specifically, we do not think so. Its evidence base lies in antidepressant selection, where trial results have shown modest effects on symptom scores and weaker effects on remission. Reviews have also flagged sponsorship and outcome-selection concerns.5 If you want the pharmacokinetic genes, sequencing or a CLIA CYP2D6 assay gives you the same information without the interpretation layer.

Does GeneSight look at ADHD meds? Its panel includes some ADHD medications, primarily atomoxetine and amphetamine-class drugs, through CYP2D6. The gene-drug pair driving those entries is the one with real guidance behind it, and it is available from any CYP2D6 genotype.

How much does genetic testing for ADHD meds cost? Commercial panels list between roughly $300 and $2,500 before insurance, with wide variation in what patients ultimately pay. A standalone CLIA CYP2D6 genotype typically costs a few hundred dollars. Calling it from whole-genome data you already have costs only compute time.

Is there a connection between MTHFR and Adderall? There is no mechanistic connection. MTHFR encodes an enzyme in folate metabolism and has no role in amphetamine pharmacokinetics or clearance. The common C677T variant affects homocysteine levels modestly and is present in a large fraction of the population. That is why professional genetics societies have recommended against ordering it routinely.

Is ADHD caused by the MTHFR gene? No. ADHD is highly polygenic, with common-variant heritability distributed across many loci and no single gene of large effect identified.3 MTHFR is not among the established ADHD loci.

Does genetic testing work for ADHD meds? For predicting whether a stimulant will help, no. For predicting atomoxetine exposure through CYP2D6, yes, and that is the boundary of what current evidence supports.6

Woolf Software builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.

Footnotes

  1. Eric Mick, Benjamin Neale, Frank A. Middleton, et al. Genome‐wide association study of response to methylphenidate in 187 children with attention‐deficit/hyperactivity disorder. American Journal of Medical Genetics Part B: Neuropsychiatric Genetics, 2008. https://doi.org/10.1002/ajmg.b.30865 ↩

  2. James J. McGough. Attention-Deficit/Hyperactivity Disorder Pharmacogenomics. Biological Psychiatry, 2005. https://doi.org/10.1016/j.biopsych.2004.10.021 ↩

  3. Lívia Balogh, Attila J. Pulay, János M. Réthelyi. Genetics in the ADHD Clinic: How Can Genetic Testing Support the Current Clinical Practice?. Frontiers in Psychology, 2022. https://doi.org/10.3389/fpsyg.2022.751041 ↩ ↩2

  4. Yilu Zhao, Zhao Fu, Eric J. Barnett, et al. Genome data based deep learning identified new genes predicting pharmacological treatment response of attention deficit hyperactivity disorder. Translational Psychiatry, 2025. https://doi.org/10.1038/s41398-025-03250-5 ↩

  5. Hays Documentation Specialists, LLC, San Mateo, CA, Priya Hays, MS, PhD. Evidence Basis for Pharmacogenetic Testing in Psychiatry. Research Review, 2022. https://doi.org/10.52845/jmrhs/2022-5-3-6 ↩ ↩2 ↩3

  6. Pratyusha Attaluri, Ayeshah G. Mohiuddin, Kowsar Teymouri, et al. Pharmacogenomics and Precision Psychiatry. Tasman’s Psychiatry, 2024. https://doi.org/10.1007/978-3-030-51366-5_130 ↩ ↩2

  7. Giulio Corrivetti, Francesco Monaco, Annarita Vignapiano, et al. Precision medicine for depression: Improving treatment response and remission. Asian Journal of Psychiatry, 2025. https://doi.org/10.1016/j.ajp.2025.104585 ↩

  8. Yichuan Liu, Hui-Qi Qu, Xiao Chang, et al. Mutation burden analysis of six common mental disorders in African Americans by whole genome sequencing. Human Molecular Genetics, 2022. https://doi.org/10.1093/hmg/ddac129 ↩