HPTEST fits the following model:
where is a function of the predictor genotype and covariates,
and is a vector of regression coefficients to be maximised over. The full likelihood is obtained as the product of the above over all samples in the data.
HPTEST currently assumes that both outcome and predictor variants are biallelic (or at least that they have been split into biallelic records before running through HPTEST). The outcome and predictor genotypes are measured with respect to those alleles, i.e. the effect is always interpreted as effect of the second predictor allele on the second outcome allele. If the outcome has ploidy larger than 1, then its alleles are regarded as forming independent draws from the Bernoulli distribution modelled above (i.e. it's a binomial logistic regression model with number of trials equal to the ploidy).
The predictor genotype is currently assumed to be diploid. To handle different possible modes of inheritance, HPTEST uses a variety of function above as follows. If denotes the predictor genotype then the following functions are used:
| Mode of inheritance | F | Specified by |
|---|---|---|
| additive | -model add |
|
| dominant | if , or otherwise | -model dom |
| recessive | if , or otherwise | -model rec |
| heterozygote | if , or otherwise | -model het |
| general | if , or otherwise | -model add+het |