A calibration currently consists of one or more named score ranges, where each range may have

• A functional class
• And/or evidence strength consisting of
  • Either an OddsPath with optional evidence strength code
  • Or evidence strength points with optional code.

However, there are score sets whose only calibration consists of functional classes not defined by score ranges. To handle such calibrations, MaveDB should be able to ingest a functional class for each variant.

• Change the calibration data model to consist of one or more named classes, where each class may have
  • A score range
  • And/or a functional class
  • And/or evidence strength consisting of
    • Either an OddsPath with optional evidence strength code
    • Or evidence strength points with optional code.
• If any class in a calibration has a score range, all classes must have score ranges. That is, the whole calibration is either range-based or not.
• Calibrations with score ranges should be defined just as they are now.
• Calibrations without scare ranges can be defined by the same means, excluding score ranges. After defining such a calibration, variants must be classified by uploading a CSV file consisting of variant identification (hgvs_nt, hgvs_pro, and/or hgvs_splice columns matching those in the scores file) and a class name. Each variant may have zero classes or one class.
  • Alternatively, we may decide to allow this to be included in the scores file. However, a separate upload is more general and better supports multiple calibrations, including third-party calibrations.
• The histogram should not display ranges for a calibration that is not range-based.
  • Alternatively, we could display ranges based on minimum and maximum scores of each class. These ranges may overlap. Whether such ranges should be displayed requires investigator input.
• Variant details should, however, still include classification when a non-range-based calibration is active.
• The classification of each variant under a non-range-based calibration may be stored in a new table.

We may allow meta-analyses to have non-range-based calibrations without scores. Such a meta-analysis would typically serve as a new functional classification of variants in an existing score set. It may make sense to develop this feature separately from support for non-range-based calibrations.