GPSeer uses three models that can be applied in serial to predict unmeasured phenotypes in genotype-phenotype maps. The linear model is always used. The classifier and spline are both optional. The classifier is always applied first, followed by the spline.
Logistic classifier that determines whether or not a genotype is below or above an assay detection threshold. Each mutation can make an additive contribution to the classifier. This uses the sklearn logistic regression model, as wrapped by the epistasis package.
A nonlinear spline to account for nonlinear mapping between the additive effects of mutations and the observed phenotypes. This uses the scipy UniverateSpline as wrapped by the epistasis package.
A linear model that describes the effects of mutations as additive perturbations to phentoype. It is possible to include pairwise and higher-ordered epistatic coefficients, however this is not recommended. This model uses the epistasis package.
Citation coming soon
Sailer & Harms (2017) Genetics.
Otwinowski et al. (2018) PNAS.