Reanalysis of Genetic Data: What Changes When the Scientific Evidence Base Is Updated

Genetic data obtained during testing does not change over time. Does that mean the interpretation of the results stays constant, and so becomes outdated and unfit for use within a few years?

Since the time of testing, new studies, larger samples, and more precise statistical methods may have appeared. They change how already obtained genetic data is interpreted. So a genetic analysis result does not necessarily stay the same simply because the DNA itself has not changed.

This applies to polygenic risk scores (PRS). They are calculated from a large number of genetic variants, using statistical weights derived from association studies. When the scientific base for such a model expands or changes, the calculation parameters may change as well. In that case, existing genetic data can be reanalyzed without collecting a new saliva sample and without repeat genotyping.

This is the basis of genetic reanalysis: the interpretation of genetic data is updated, while the person's genetic profile itself stays the same.

The Genotype Is Fixed, the PRS Model Can Be Updated

A person's genotype is the set of variants at specific positions in the genome. It is determined once, when the sample is genotyped, and this data does not change over time.

What changes is how this data is converted into the final score: that calculation is performed by the PRS model, which is reviewed when the results of new studies are incorporated into its construction and further validation. The outcome of such a review depends on the sample the model was developed on, the method it was built with, and how it was validated — these parameters are among the mandatory elements of the description of any polygenic score (Wand et al., Nature, 2021).

To learn why genetic data doesn't go out of date over time, see the article Genetic Data Doesn't Go Out of Date: Why It's Worth Returning to Your Report.

Variant Weight: The Result of a GWAS Statistical Analysis

The weight of each variant is determined by a statistical analysis of the association between that variant and a trait within a large cohort of participants in a genome-wide association study (GWAS). These weights are one of the parameters that determine the model's output.

Large samples increase the statistical power of a GWAS: as the sample size grows, a study can detect more genetic associations and estimate variant effects more precisely (Uffelmann et al., Nat. Rev. Methods Primers, 2021). Such samples come from large-scale biobanks — in particular the UK Biobank, a resource that combines genetic data with detailed phenotyping of hundreds of thousands of participants (Bycroft et al., Nature, 2018).

Several different PRS models can exist for a single trait, built on different samples and with different computational methods. Open catalogs, including the Polygenic Score Catalog and GWAS Catalogue, collect published models along with their variants, weights, and metadata, and make it possible to compare them with one another (Lambert et al., Nature Genetics, 2021).

Three Components of a PRS Model Update

A PRS model update covers three independent types of change: new weights for existing effects, a change in the set of effects (adding or excluding certain variants), and a change in the reference panel. These changes affect different components of the calculation, so their consequences differ:

• refinement of variant effect estimates — when the results of a new GWAS are used to estimate effects with greater statistical precision

• a change in the composition of the PRS model — when newly associated variants are included in a new version of the model

• a change in the reference sample — when the result is standardized against a different population or cohort.

Refining effect estimates and changing the model's composition both concern the model itself: they change how genetic data is converted into the final score if the new estimates or variants are included in its updated version. A change of reference sample concerns what the result is compared against: it changes the percentile or standardized value calculated relative to the new sample, while the calculation of the weights themselves stays the same.

The scale to which the set of associated variants can expand is seen in the example of height — a trait with a large body of accumulated data. A GWAS of 5.4 million participants identified 12,111 independent variants associated with height. Together they explain nearly all of the trait's heritability attributable to common variants (Yengo et al., Nature, 2022).

A Model Update Refines the Earlier Result

The fact that a model has been updated does not mean the earlier result was wrong. It was calculated according to the methodology and data available at a particular point in time. After an update, a different set of data or methodological parameters may be used for the same genetic information.

At the same time, a PRS in any version of the model remains a measure of genetic predisposition, not a diagnosis. Its standardized value or percentile makes it possible to assess a person's position relative to the reference sample. It does not establish that a particular condition will necessarily develop (Slunecka et al., Hum. Genom., 2021). What gets updated is the calculation and the way the result is interpreted, not the genetic data it is based on.

It is also worth separating genetic predisposition from current physiological state. A polygenic score describes baseline predisposition calculated from DNA variants and complements clinical and laboratory data that reflect the state of the body at a specific point in time.

Update Frequency Depends on the Trait

The pace of updates differs across traits. Traits with active research and growing samples give more grounds for revising a model. Whether a revision actually happens, however, depends on the methodology and practice of model development.

The scale of the change also depends on what exactly was updated: the weights, the composition of variants, or the normalization parameters. Such a change reflects the dependence of a PRS on the data and methodology the model is built on.

The Apixmed Prism genetic test lets you explore individual characteristics based on DNA — see details in the DNA test catalog.

Updating the Interpretation Doesn't Mean Retesting

An updated interpretation is not a reason to doubt the original test result. The science of polygenic traits works with probabilities, which are refined as evidence accumulates. Genetic data obtained during testing remains the basis for further interpretation. At Apixmed Prism, indicators can be reanalyzed without collecting a new sample. For this, you can use an existing genetic file (VCF), including one obtained from another company.

To learn more about how to work with your report next — reading the indicators, setting priorities, knowing when to see a doctor — see the article What to Do With Your Report Next: From Data to Decisions.

Frequently Asked Questions

What exactly can be updated in a genetic report?

The effect estimates of individual variants, the composition of the model used to calculate the polygenic score, and the reference sample against which the result is standardized. The person's set of genetic variants remains the same.

Where does the data used to calculate a PRS come from?

From the results of genome-wide association studies carried out by independent scientific consortia on large cohorts of participants, as well as from open catalogs that collect published models and their parameters.

Can I compare my score with the score of someone who took the test at a different time?

A direct comparison of two numbers calculated with different versions of the model is not valid. A score only makes sense in the context of the model and reference sample it was calculated with.

Genetic test results are not a diagnosis and not a substitute for a consultation with a doctor. The Apixmed Prism report provides genetic context that complements examination results and helps you make decisions together with your doctor. 

Sources

1. 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

2. Uffelmann, E., Huang, Q. Q., Munung, N. S., de Vries, J., Okada, Y., Martin, A. R., Martin, H. C., Lappalainen, T., & Posthuma, D. (2021). Genome-wide association studies. Nature Reviews Methods Primers, 1(1), 59. https://doi.org/10.1038/s43586-021-00056-9

3. Bycroft, C., Freeman, C., Petkova, D. et al. (2018). The UK Biobank resource with deep phenotyping and genomic data. Nature, 562(7726), 203–209. https://doi.org/10.1038/s41586-018-0579-y

4. 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

5. Yengo, L., Vedantam, S., Marouli, E. et al. (2022). A saturated map of common genetic variants associated with human height. Nature, 610(7933), 704–712. https://doi.org/10.1038/s41586-022-05275-y

6. 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, 46. https://doi.org/10.1186/s40246-021-00339-y