GET-Evidence: Prioritization Score

Prioritization scores

Every genome has millions of variants, thousands of which are potentially interesting. To help sort through these and prioritize variants for review, an “prioritization” is automatically calculated. This score combines two different reasons for prioritizing variants — we want to prioritize variants which we think have some relevant literature, and we want to prioritize variants for which computational methods predict a damaging effect.

How the prioritization score is calculated

Prioritization scores can be a total of six points, with a maximum of two points coming from each of three different categories: (1) pre-existing variant-specific knowledge, (2) pre-existing gene-specific knowledge, (3) computational prediction of variant disruptiveness.

Pre-existing variant-specific knowledge

Presence in variant specific databases and potential or confirmed webpage matches contributes to our score for pre-existing variant-specific knowledge. This component of score reflects our desire to prioritize variants which have some relevant published information for review.

  • +2 for presence in Online Mendelian Inheritance in Man (OMIM) – this database is given more points than other sources because it tends to contain rare, severe disease variants — of course, it also has benign and milder variants, users must look at OMIM itself to find out more.
  • +1 for presence in the Pharmacogenomics Knowledge Base (PharmGKB) – this database includes “susceptibility” variants and variants that may affect drug metabolism
  • +1 for presence in the Human Genome Epidemiology Network (HuGENet) database – this database contains many “susceptibility” variants that come from genome-wide association studies (GWAS).
  • +1 if any unevaluated or confirmed positive webpage hits are present (+0 if all webpage hits are either confirmed false or have a tied vote)

Although these can add up to more than two, this score has a cut-off maximum of two points.

Pre-existing gene-specific knowledge

Not all genes are linked to human disease and phenotypes — this part of the score prioritizes review for variants occurring in genes believed to have a phenotype impact. We use the GeneTests database, which contains data about genetic diseases and gene testing.

  • 1 point if a gene has clinical testing available
  • 2 points if a gene has clinical testing available and is linked to a GeneReviews article about a genetic disease

Computational prediction of variant disruptiveness

Some variants might be predicted to cause disease even though they have not been reported upon specifically in the literature. This part of the score prioritizes variants where there is some prediction of disruptive effect that exists, independent of publications. For this, we use Polyphen 2, a tool which predicts functional effects for single nucleotide substitutions.

  • 1 point if Polyphen 2 predicts “possibly damaging” (score is < 0.85 and >= 0.2)
  • 2 points if Polyphen 2 predicts “probably damaging” (score is >= 0.85)
  • 1 point for a multiple amino acid substitution, in-frame insertion or deletion, or Polyphen 2 score is otherwise unknown.
  • 2 points for a nonsense or frameshift mutation

In addition, if this part of the score is nonzero and the allele frequency of the variant is 5% or greater (in GET-Evidence’s set of PGP and public genomes), 1 point is subtracted.

What the prioritization score is not

Prioritization scores are not a prediction of a variant as having an effect — Although this is a component of the score, the score is intended to prioritize variants for review. There may be variants with no functional effect that are highly reported-upon, thereby getting a higher score. There may be undiscovered variants with severe effect.

Prioritization score is not a summary of a variant’s evaluation — Although voting for web hits can adjust the score, a prioritization score merely reflects whether a variant is prioritized for review. It does not change in response to most aspects of variant review (users classifying a variant as pathogenic or benign, or adding variant impact score data).

Gene search

"GENE" or "GENE A123C":

Log in