Document genes that shouldn't be set to reference

[markwoon]

We should document variants in certain genes that shouldn't be set to reference (if missing, as opposed to actually reference):

• CYP3A5
• CYP2C19
• UGT1A1

[katrinsangkuhl]

This affects specific positions within those genes:

If the entire datasets are systematically missing the following variants:
CYP2C19-rs3758581 (this position is the definition of *1)
If "set to reference" but truly unknown (as compared to being reference) might result in a high number of "not called" for CYP2C19 given this position is included as variant in the majority of CYP2C19 allele definitions https://www.pharmvar.org/gene/CYP2C19

UGT1A1-rs3064744 (this position (TA repeat) is the definition of *28, *37, *36)
If "set to reference" but truly unknown (as compared to being reference) might result in a higher number of *80 genotypes which results in an Indeterminate phenotype. *80 and *28 (*37) are in a high linkages and CPIC definitions include *28, *37, *80+*28, and *80+*37 as decreased function alleles and *80 as unknown function allele.

CYP3A5-rs776746 (this position is the definition of *3)
If "set to reference" but truly unknown (as compared to being reference) will result in *1/*1 in case no other variant is present. CYP3A5*3 is a quite common variant, for frequency data see https://www.pharmgkb.org/haplotype/PA166128219.
PharmCAT will report the sample also as *1/*1 if rs776746 is missing and no other variant is present but include a warning that rs776746 should be assayed to more accurately determine the CYP3A5 phenotype.

[katrinsangkuhl]

As I put in slack, If its just in a couple of samples because for whatever reason a variant is missing I think it as good as setting other variants to reference and getting *1, I am thinking more about entire datasets that miss the variant systematically.