Skip to content

Classification trees

In order to end up with a small list of candidate variant records for interpretation VIP performs variant filtration by:

  1. Classify all variant-consequences based on variant annotations
  2. Remove variant-consequences based on their classes
  3. Annotate remaining variant records using inheritance matcher
  4. Classify all variant-consequences based on variant annotations in the context of samples
  5. Remove variant-consequences based on their classes.
  6. Remove variants that had all their variant-consequences removed

The following sections describe the default variant filtration strategies and how to customize classification and filtration.

Default

VIP contains default filtration strategies for variant-consequences (GRCh37 and GRCh38) as well as variant-consequences in the context of samples.

Variant-consequences

The default decision tree to classify variant-consequences works as follows:

  1. Each variant-consequence is classified as Benign, Likely Benign, VUS, Likely Pathogenic, Pathogenic or Remove
  2. Variant-consequences classified as Benign, Likely Benign and Remove are removed

Default classification tree

Above: default GRCh38 variant classification tree

Variant-consequences (samples)

The default decision tree to classify variant-consequences in the context of samples works as follows:

  1. Each variant-consequence-sample is classified as OK, LQ (low quality) or MV (mendelian violation)
  2. Variant-consequences classified as LQ for all samples are removed

Default variant-sample classification tree

Above: default variant sample classification tree

Customization

The default variant filtration strategy can be customized using the following parameters (see here):

  • vcf.classify.GRCh37.decision_tree
  • vcf.classify.GRCh38.decision_tree
  • vcf.filter.classes
  • vcf.classify_samples.GRCh37.decision_tree
  • vcf.classify_samples.GRCh38.decision_tree
  • vcf.filter_samples.classes

The following repositories might be of interest when creating a new decision tree:

You are free to use your own set of classes in your decision tree. Keep in mind to update the filter classes parameters accordingly.