Genetic Test Accuracy: DNA Reading and Risk Assessment

A genotyping-based genetic test determines common DNA variants reliably: compared with direct DNA sequencing (as in whole-genome sequencing), the concordance of results is about 99%. Predicting whether a particular clinical condition will develop from information about these common variants is a different task with its own, narrower limits.
Reading accuracy and prediction accuracy do not coincide because of the structure of polygenic traits: overall predisposition is made up of many small contributions, so a prediction for an individual depends on the trait, the model, and the comparison group.
So what stands behind the accuracy of a genetic test, and what should you look at in a specific report?
Two Components of Genetic Test Accuracy
To obtain a score for a metric, two different steps are performed: a person's DNA variants are determined, and then, from those variants, the person's relative predisposition to a given condition is calculated. Each step has its own type of error, and each has its own measure: analytical validity for variant calling, and clinical validity for the calculation of predisposition.
Analytical validity describes the quality of laboratory variant calling: whether the genotypes in the report match the original genome of the sample. It is tested by comparing genotyping results with sequencing of the same samples — a method that reads the DNA sequence directly rather than at predefined positions in the DNA.
Clinical validity shows how well the predisposition estimate agrees with what is actually observed in people: whether the condition occurs more often in those with higher scores, and whether the expected frequency matches the actual one. The suitability of an estimate for a specific medical decision lies beyond validity and is assessed separately.
High analytical validity does not guarantee high clinical validity: reliable reading, although it provides accurate data, has only an indirect effect on the accuracy of the risk estimate.
Accuracy of DNA Reading
In a study using UK Biobank data, genotyping results were compared with exome sequencing for 49,908 participants (Weedon et al., BMJ, 2021). For common variants with a population frequency above 1%, genotyping correctly detected 99.7–99.8% of variants. Among all variants called as present, sequencing confirmed concordance in 98.7–99.0% (Weedon et al., BMJ, 2021).
For rare variants, the same study gives different numbers. Among variants with a frequency below 0.001%, sequencing confirmed only 16% of the 4,757 results in which genotyping indicated the presence of the variant. Separately, the authors assessed rare pathogenic variants using two genes as an example and concluded that for very rare pathogenic variants such results are unreliable and should not influence medical decisions without confirmation by an alternative clinically validated method (Weedon et al., BMJ, 2021).
Importantly, some genotypes are determined computationally: from the positions that were read and a reference panel of haplotypes (sets of variants inherited together on one stretch of a chromosome). This is how the likely genotype is estimated where direct reading does not provide it. Such computation expands the range of variants available for research — with large panels down to a frequency of about 0.01% (Taliun et al., Nature, 2021). For an individual person, this does not mean equal accuracy in determining each of these variants.
For more on how genotyping data becomes report metrics, see the article From Raw Data to Report Metrics: How Genotyping Data Becomes a Genetic Profile.

Why Accurate Reading Does Not Mean an Accurate Risk Estimate
Many traits and conditions assessed in genetic reports have a polygenic architecture: overall genetic predisposition is formed by the contributions of hundreds or thousands of variants, each with a small individual effect. To express this estimate as a single value, a polygenic risk score (PRS) is calculated, which determines a person's level of genetic predisposition relative to a reference population (Slunecka et al., Hum. Genom., 2021). PRS is built on the results of genome-wide association studies (GWAS), in which variant frequencies are compared between groups with and without the trait.
A high polygenic risk score (PRS) does not in itself mean a high probability of falling ill. For example, a PRS at the 90th percentile means that roughly 90% of people in the reference sample have a lower genetic predisposition score. To assess absolute risk, you also need to take into account how often the disease occurs in the relevant population over a given period. Depending on the disease, age, living conditions, and clinical indicators may matter for an individual prediction. How well does a PRS distinguish people with the disease from those without it? This is shown by the area under the ROC curve (AUC). Imagine 100 randomly formed pairs of people, in each of which one person has the disease and the other does not. If the AUC equals 0.70, then in about 70 of the 100 pairs the person with the disease will have the higher genetic predisposition score. With an AUC of 0.50 the result is the same as random selection, and with an AUC of 1 the person with the disease will have the higher score in all 100 pairs.
In a 2023 review of the clinical utility of polygenic scores, this indicator was typically 0.65–0.70 and depended substantially on the disease; for type 1 diabetes the authors report about 0.90. Adding PRS to a clinical scale raised it only moderately. A high PRS is not equivalent to a high absolute risk (Koch et al., J. Community Genet., 2023).
The reason for this limitation lies in the architecture of the trait itself. The genetic component is only one of several factors alongside lifestyle, environment, age, and chance, and its weight differs between conditions. High accuracy in determining common variants does not change the fact that genetic data describe only part of the variation in a trait.
Accuracy of the Predisposition Estimate
For a single trait or disease there are several calculation models, built on different samples and with different statistical methods, so the result depends on the choice of model. These models are also described inconsistently, and to bring the description to common requirements, a reporting standard for polygenic risk scores (polygenic risk score reporting standards, PRS-RS) has emerged in the field. The standard was developed by a working group of the consortium Clinical Genome Resource (ClinGen) together with the authors of the Polygenic Score Catalog (PGS Catalog). The standard contains 22 items and defines the mandatory information about each model: sample composition, statistical methods, validation results, and limitations of the estimate (Wand et al., Nature, 2021). The catalog stores the models themselves, with lists of variants, alleles, and weights — numerical coefficients that set the contribution of each variant — so that the models can be reproduced and compared with one another (Lambert et al., Nat. Genet., 2021).
Accuracy also depends on the population. The greater the genetic distance between the population whose data the model was built on and the one to which it is applied, the lower the prediction accuracy. Added to this are differences in variant frequencies and in the share of trait variation explained by genetic factors (Kachuri et al., Nat. Rev. Genet., 2024).
Apixmed Prism reads about 700,000 genetic markers and calculates each metric using all available models for the trait, bringing their results to a common scale relative to a reference population.

Limits of Genetic Assessment
Genetic assessment covers relative predisposition and is not a clinical diagnosis. It does not show the current state of the body: levels of substances, inflammation, and organ function are assessed by laboratory tests. Therefore, a genetic result is read together with the patient's history and examination data, and interpretation in a medical context requires a specialist with appropriate training (ACMG, DNA-based screening and personal health, 2021).
Confusing genetic predisposition with the current state is the most common source of wrong expectations of a test.
For how these two levels of information relate, see the article What a DNA Test Is and How It Differs from a Blood Test.
Variant calling and prediction have different limits: for common variants, reading is reliable, while the strength of prediction depends on the trait and the model. Both limits are assessed by defined criteria, so the reliability of a result is judged not by the number itself: you need to know by which method the data were read, by which model the estimate was calculated, and relative to which population it is presented.
Explore your genetic traits with Apixmed Prism — more than 1000 metrics in 26 areas of genetic testing.
Frequently Asked Questions
Can DNA variant calling contain errors?
For common genetic variants the probability of error is low: comparison with direct reading of the DNA sequence, that is, sequencing, gives an accuracy of about 99%. For very rare variants, reliability drops, so such results are confirmed with a targeted laboratory method.
A genetic test showed a high level of predisposition, but standard examinations found no abnormalities — how should this be understood?
These results describe different things. A genetic report shows inherited predisposition, while standard laboratory tests and examinations show the state of the body at the time of the study. They do not have to agree, and a discrepancy does not indicate an error in either result.
Does a high genetic predisposition mean the condition will develop?
A high predisposition is not equivalent to a high absolute risk. A PRS value shows how much a person differs in genetic factors from most people in the reference sample. Whether the predisposition manifests depends on many other circumstances.
Do genetic tests become more accurate over time?
The inherited genotype remains unchanged, while calculation models are revised as research accumulates, so reinterpretation of the same data may give a refined estimate. More details are in the article Report Reanalysis: What Changes When the Scientific Evidence Base Is Updated.
The results of a genetic test are not a diagnosis and do not replace a doctor's consultation. The Apixmed Prism report provides genetic context that complements examination results and helps you make decisions together with your doctor.
Sources
-
American College of Medical Genetics and Genomics. (2021). DNA-based screening and personal health: a points to consider statement for individuals and health-care providers. Genetics in Medicine, 23(6), 979–988. https://doi.org/10.1038/s41436-020-01083-9
-
Kachuri, L., Chatterjee, N., Hirbo, J. et al. (2024). Principles and methods for transferring polygenic risk scores across global populations. Nature Reviews Genetics, 25(1), 8–25. https://doi.org/10.1038/s41576-023-00637-2
-
Koch, S., Schmidtke, J., Krawczak, M., Caliebe, A. (2023). Clinical utility of polygenic risk scores: a critical 2023 appraisal. Journal of Community Genetics, 14(5), 471–487. https://doi.org/10.1007/s12687-023-00645-z
-
Lambert, S. A., Gil, L., Jupp, S. et al. (2021). The Polygenic Score Catalog as an open database for reproducibility and systematic evaluation. Nature Genetics, 53(4), 420–425. https://doi.org/10.1038/s41588-021-00783-5
-
Slunecka, J. L., van der Zee, M. D., Beck, J. J. et al. (2021). Implementation and implications for polygenic risk scores in healthcare. Human Genomics, 15(1), 46. https://doi.org/10.1186/s40246-021-00339-y
-
Taliun, D., Harris, D. N., Kessler, M. D. et al. (2021). Sequencing of 53,831 diverse genomes from the NHLBI TOPMed Program. Nature, 590(7845), 290–299. https://doi.org/10.1038/s41586-021-03205-y
-
Wand, H., Lambert, S. A., Tamburro, C. et al. (2021). Improving reporting standards for polygenic scores in risk prediction studies. Nature, 591(7849), 211–219. https://doi.org/10.1038/s41586-021-03243-6
-
Weedon, M. N., Jackson, L., Harrison, J. W. et al. (2021). Use of SNP chips to detect rare pathogenic variants: retrospective, population based diagnostic evaluation. BMJ, 372, n214. https://doi.org/10.1136/bmj.n214













