xiamaz
commented
2 years ago I will give some examples based on the ACGS Guidelines for ACMG Classification based on the 2019 and 2020 version.
Variable weight of criteria
The PVS1 criteria can be applied at either very strong, strong or moderate depending on variant type, location or evidence for the likelyhood of a null effect.
The PS2 criteria can be applied from supporting up to very strong on a score based scale.
The PS4 criteria can be applied at moderate if the same variant has been identified in multiple affected unrelated individuals, or at supporting if only one.
BS1 can be used as standalone for AD disorder with high penetrance.
Valid combination of criteria
PVS1 should not be used in combination with PP3 or PM4. These are mutually exclusive.
BP3 should not be used in combination with BP4
Some rules can be substituted with other rules if evidence is not sufficient. These are not mutually exclusive, but selection of both criteria might be a user error. An example is instead of PS2, PM6 can be used if identity is assumed but not confirmed.
Ruleset requirements
The previous examples have no claim to being comprehensive and should just serve as a point of reference for possible specifications in the ACGS guidelines.
From these an implementation for a ruleset should allow for:
- versioning
- valid range of weight for each criteria
- forbidden combinations of criteria
- possibly invalid combinations of criteria
Additionally it would be nice to have:
- hints for individual weights
- hints if forbidden combinations are used
- hints if combinations might make no sense (for example PM6 might be used instead of PS2, but the combination is not forbidden if two lines of evidence do exist, in this scenario it would be great if the user could be notified that this might be a mistake)
holtgrewe
Is your feature request related to a problem? Please describe. Grading using the ACMG criteria and scheme is very useful. As also mentioned in #476, specific groups and labs have made their own adjustments. Different schemes might be be used even at the same site, for example when used for ultra-rare vs. "just rare" vs. predisposition/risk grading. We need to cater to that.
Describe the solution you'd like We keep the list of ACMG criteria fixed and also the list of categories we also keep the existing names.
We introduce an
AcmgBasedSchemamodel (or similar) that captures the information which criteria goes into which category. For each such instance we also record a list of pairwise incompatible settings for simple sanity checking (aka "linting). We store a name and a version and also whether it is a "factory preset".These schemas can be curated by a VarFish administrator only at the moment but assigned to projects by the project owner and to cases by contributors.
A migration creates an entry here that is named "ACMG 2015" in "version 1" and is a factory preset. All existing gradings are converted to use this.
Each case is annotated with which schema and version is used, each project defines the default grading schema. When a case is opened, users can adjust the ACMG based schema. In case of existing assessments, they get a preview of all graded variants and how changing the schema would change the assessment. They can either chose to apply the change (no way to revert right now), not apply the change and keep the schema constant, or clear all ACMG criteria assessments for now (with an additional confirmation).
We should also allow users to easily create a PDF document with the used settings that can be used/provided for documenting the assessment in one place and, e.g., be published on a website. Something like this is required for ClinVar submission anyway.
Describe alternatives you've considered N/A
Additional context
202