Investigate low scoring diagnostic hits

[damiansm]

Look at the known diagnoses where variant is part of Exomiser results but outside top 5 and assess what would improve the ranks for these i.e. better pheno-matching algorithms, better underlying phenotype data or better variant scoring

[damiansm]

Assessed the 20/334 pilot diagnoses that are in Exomiser results but outside top 5:

• 8 involve a low scoring variant (including 4 that have low scoring phenotype match as well)

  • 3 due to issues with b37 MT assembly on wrong strand that is fixed in b38
  • 2 involved the known pathogenic Clinvar variant 16:89613145:C:T in SPG7 that is pretty common in populations, especially GEL local frequencies (almost 1%)
  • 1 involved the known pathogenic Clinvar variant 13:20763685:AC:A in GJB2 that is pretty common as well
  • 1 involved the known pathogenic Clinvar variant 12:120878246:TCACTC:T in COX6A1 that is a splice region variant so gets a lowish default score
  • 1 is really an issue with incomplete penetrance. The variant on its own does not pass strict inheritance and Exomiser results are for the low-scoring cmphet

• 16 involved a low scoring phenotype match:

  • 2 missed because seizures (HP:0001250) is now below IC 2.5 cutoff in db - see other issue
  • 1 has no real match between the disease and patient phenotypes
  • 2 look like discrepanices between expected MOI from GMC/GeL/panelApp and OMIM/Orphanet dbs
  • Rest match ok'ish but just score low as common phenotypes and not all terms matched

On whole looks like little evidence that better phenotype or variant scorers would really fix any of this. Always considering known pathogenic Clinvar variants as top candidates would help.