How to filter large CBU search result sets by Predicted Match Counts?

[zabeen]

The Problem

The latest recommendations for CBU search are to run 4/8 search at the allele-level. Allowing for 4 mismatches at any locus can lead to 10s of thousands of CBUs being returned, which is not a manageable amount for end-users.

[zabeen]

Proposed solution

Atlas generates a predict match count per donor/CBU as part of calculating match probabilities. This is the sum total of positions that have been deemed as Exact or Potential matches, based on the locus match probabilities.

The CBU result set returned by Atlas could be be filtered by predicted match count, to hide units that have too many predicted mismatches. This filtering step should happen at some point before the results are displayed to the user, but not within Atlas itself.

E.g., a CBU was called a 5/8 by the matching algorithm, but only has a predicted match count of 2/8. This unit should not be displayed to the end-user, by default.

In addition, the remaining CBUs could be ranked to promote those units that have the fewest predicted mismatches, e.g., order units by:

• Match count (as determined by the matching component), descending
• Predicted match count, descending
• Total Nucleated Cell (TNC) count, descending

Testing

In a CBU search run during internal testing, a match list of 46,056 units was reduced to 13,412 by applying such a filter.

Additional Requirements

• "Unrepresented" CBUs (those that do not have match probabilities calculated as they cannot be explained by the referenced haplotype frequency (HF) dataset) should not be filtered out.
• There should be confidence in the HF datasets from which predictions are made, and in how the datasets are mapped to CBU ethnic group types.
• Ideally users should have the option to see the unfiltered CBU match list, e.g., by toggling the filter option in the search user interface.