Transformation of phenotype component by applying a user-specified non-linear transformation to the phenotype component.

transformNonlinear(component, alpha, method, logbase = 10, power = 2,
  expbase = NULL, transformNeg = "abs", f = NULL, verbose = TRUE)

Arguments

component

[N x P] Phenotype matrix [double] where [N] are the number of samples and P the number of phenotypes

alpha

[double] weighting scalar for non-linearity: alpha==0 fully linear phenotype, alpha==1 fully non-linear phenotype. See @details.

method

[string] one of exp (exponential), log (logarithm), poly (polynomial), sqrt (squareroot) or custom (user-supplied function)

logbase

[int] when method==log, sets the log base for transformation

power

[double] when method==poly, sets the power to raise to.

expbase

[double] when method==exp, sets the exp base for transformation.

transformNeg

[string] one of abs (absolute value) or set0 (set all negative values to zero). If method==log and transformNeg==set0, negative values set to 1e-5

f

[function] function accepting component as a single argument.

verbose

[boolean]; If TRUE, progress info is printed to standard out.

Value

[N x P] transformed phenotype matrix [double]

Details

transformNonlinear takes a phenotype component as input and transforms it according to the specified transformation method. The user can choose how strongly non-linear the resulting phenotype component should be, by specifying the weighting parameter alpha: component_transformed = (1 - alpha) \* component + alpha \* transformfunction(component)

Examples

# Simulate non-genetic covariate effects cov_effects <- noiseFixedEffects(N=100, P=5) # Transform logarithmically covs_log <- transformNonlinear(cov_effects$shared, alpha=0.5, method="log", transformNeg="abs")
#> Use log as transformation method
# Transform custom f_custom <- function(x) {x^2 + 3*x} covs_custom <- transformNonlinear(cov_effects$shared, alpha=0.5, method="custom", f=f_custom)
#> Use custom as transformation method