RAREMETAL
Bug fixes:
-
Fix two critical bugs that cause covariance values to stop being read (and left to their default of 0) when encountering missing or excluded variants
-
If a variant is missing in one study, present in another, and is in a group (list of variants for a gene) - whenever this variant is encountered while loading the covariance values for the group (in the study missing the variant), instead of skipping it and continuing, the remaining covariance values for that group will be skipped. (0e8800f, 71f85fb)
-
If a variant fails one of the following filters in one study but not others:
- HWE or call rate filters
- Mismatches alleles from a previous study
- Missing an allele
The same issue as in (i) will occur. In the study in which the variant fails QC, while attempting to read the covariance value for that variant in the group being tested, the program will fail to find the value on the row in the covariance matrix, and return early, rather than continuing to load the remaining covariance values for that group. (344c811, de92d92).
These bugs can result in covariance values being set to 0 when they should not be, which can lead to false positive associations.
For the majority of rare variant tests, the covariance between variants is typically quite small, and therefore being truncated to 0 will likely not affect results. However, it is entirely possible that for certain masks, with more common variants included, that the covariance would not be close to 0. In these cases, a false positive association could occur.
These bugs appear to have been introduced in v4.14.1 (75a7e98) on Oct 6, 2017.
-