Genetics versus genomics
Genetics is the study of individual genes and how traits are inherited — the classic single-gene conditions such as cystic fibrosis, Huntington's disease or sickle cell disease, where one gene largely determines the outcome. Genomics is broader: the whole genome, how genes interact, how they are regulated, and how variation across thousands of positions contributes small amounts to common traits and risks. Most common conditions — heart disease, type 2 diabetes, depression, most cancers — are genomic rather than genetic in this sense, involving many variants each contributing a little, against a large background of environment and chance.
The distinction matters because it determines how useful a test can be. For a single-gene condition, a genetic test can give a clear answer with real clinical consequences. For a common complex condition, no test can tell you what will happen, because genetics is one input among many. This is the central misunderstanding in consumer genomics: people expect the clarity of the first from a test that is measuring the second.
Pharmacogenomics: the part already in use
Pharmacogenomics is the most clinically mature application, because the question is narrow and the action is clear: will this person metabolise or react to this drug unusually. Several gene-drug pairs are used in routine practice. Testing for the HLA-B*57:01 variant before starting abacavir for HIV substantially reduces hypersensitivity reactions and is standard of care. DPYD variants predict severe, sometimes fatal toxicity from fluoropyrimidine chemotherapy such as 5-fluorouracil and capecitabine. TPMT and NUDT15 variants predict severe myelosuppression from thiopurines including azathioprine and mercaptopurine.
Others are established but applied more variably. CYP2C19 status affects the activation of clopidogrel, with implications after coronary stenting; CYP2D6 status affects codeine, which relies on conversion to morphine, and is relevant to some other drugs. HLA-B*15:02 is associated with severe skin reactions to carbamazepine in people of certain Asian ancestries, and testing is recommended in those groups. G6PD deficiency, long known, changes the safety of several drugs. What these share is a specific gene, a specific drug, a specific consequence and a defined alternative — which is exactly what broad wellness-oriented pharmacogenomic panels usually lack. If you are offered testing, ask which drug decision it will change.
Polygenic risk scores and their current limits
A polygenic risk score adds up the small effects of many common variants to produce a single number describing genetic predisposition to a condition. At population scale these scores work: people in the top few per cent of a well-constructed score for coronary artery disease genuinely have higher average risk than those at the bottom. That is a real scientific achievement and it is being actively studied for clinical use.
The problems appear when the score is applied to one person. The distributions overlap heavily, so most people sit in a middle where the score changes very little about what they should do. The scores explain a modest share of total risk for most conditions, with conventional factors — blood pressure, lipids, smoking, weight, activity, family history — often mattering more and being modifiable, which genes are not. Crucially, scores transfer poorly across ancestries: a score derived predominantly in European-ancestry populations loses accuracy in others, sometimes substantially. And for most conditions there is no trial evidence that giving someone their score improves any outcome. The reasonable current position is that these are promising research tools whose clinical role is still being defined.
Direct-to-consumer testing: what to know first
Consumer genetic tests are genuinely useful for some things and systematically over-interpreted for others. The most important technical point is coverage: many consumer tests genotype a selected set of variants rather than sequencing a gene. A test that examines three BRCA variants common in one population will miss the great majority of pathogenic BRCA variants, so a negative result is not reassurance. Anyone with a family history suggestive of hereditary cancer needs clinical genetic testing with counselling, not a consumer kit.
Second, results without counselling are hard to interpret. Variants of uncertain significance are common, and their meaning changes as databases grow. Third, the raw-data files these companies provide, run through third-party interpretation sites, produce a large number of false positives — the underlying genotyping was never validated for clinical use. Fourth, privacy: genetic data is not covered by clinical privacy law when you buy the test yourself, it describes your relatives as well as you, and it cannot be changed after a breach. Ancestry and trait reports are entertainment and mostly harmless. Health reports deserve to be treated as a prompt to talk to a clinician rather than an answer.
Where it is furthest along: cancer
Oncology is the clearest working example of personalised medicine, because tumours can be profiled directly and the results change treatment. Testing tumour tissue for specific alterations routinely determines therapy in several cancers — HER2 status in breast cancer, EGFR and ALK alterations in lung cancer, BRAF in melanoma, and mismatch-repair deficiency or high microsatellite instability as a marker across tumour types. Some approvals are now defined by the molecular alteration rather than the organ the cancer started in, which is a genuine conceptual shift.
Germline testing also matters here: inherited BRCA1 and BRCA2 variants and Lynch syndrome change screening, prevention and sometimes treatment for the patient and for relatives who can then be tested. The honest limits are that only a minority of patients have an actionable alteration, that resistance frequently develops, and that access to both testing and the resulting drugs is uneven. But this is where the promise has most clearly become practice.
Who the reference data leaves out
Most large-scale genomic reference datasets still over-represent people of European ancestry by a wide margin. This is not a minor sampling issue; it degrades the technology for everyone else. Variant interpretation depends on knowing how common a variant is in a person's population, so variants of uncertain significance are returned more often for people of underrepresented ancestries, and misclassification has occurred where a variant common and harmless in one population was flagged as pathogenic based on its rarity in another.
Polygenic risk scores are affected most, losing meaningful accuracy when applied outside the population they were derived in. Pharmacogenomic frequencies also differ substantially between populations, which is why some testing recommendations are ancestry-specific. Deliberate efforts to broaden participation are underway and the imbalance has narrowed, but it has not gone away, and it is a fair question to ask of any genomic test: which populations was this validated in. Our research hub covers why representative data is a scientific issue and not only a fairness one.