Version History

We recommend that all users upgrade to v1.12 to obtain the most accurate results with sparse data sets.

• v1.12: Changed permutation p-value calculations to perform approximate floating-point comparisons between permutation and observed test statistics using a user-specified relative epsilon. This change increases the robustness of these comparisons to normal floating-point rounding and truncation errors and therefore yields more accurate permutation p-values when the permutation null distribution is highly discrete. To help identify such discrete permutation null distributions, the output now also includes an estimate of the permutation null distribution probability mass function at the observed test statistic value.
• v1.11: Modified hotellingt2 function used in CMC to check whether F statistic numerator df (i.e., rank of S) are > 0 and return -99 for all results variables if not. This prevents premature program termination due to GSL domain errors when the rank of S is 0.
• v1.10: Changes to improve robustness in small, highly selected samples with higher rates of missing genotypes. Specifically:
  1. The program now excludes variants with only a single genotype (0, 1, or 2 minor alleles) observed in the sample to avoid NaN CA statistics. Previously, only variants with all 0 genotypes in the sample were excluded.
  2. The program now issues warning messages to STDERR if any variants have observed genotypes only in cases or controls in the observed sample or in any permutation. This can only happen if there are as many or more individuals with missing genotypes in the entire sample than there are individuals in either the case or control groups.
• v1.9: Initial publicly released version.