Convert a periodic time variable into components usable in linear models

Decompose a periodic time variable into multiple components based on either the first harmonic of a Fourier series or on a periodic smoothing spline.

Usage

limorhyde(
  time,
  colnamePrefix = NULL,
  period = 24,
  sinusoid = TRUE,
  nKnots = 3,
  intercept = FALSE
)

Arguments

time: Numeric vector of times, e.g., at which samples were acquired.
colnamePrefix: Character string with which to prefix the column names of the basis.
period: Number corresponding to the period to use for the decomposition (in the same units as time).
sinusoid: If TRUE, the decomposition is based on cosinor, i.e., cosine and sine. If FALSE, the decomposition is based on a periodic smoothing spline from the pbs package.
nKnots: Number of internal knots for the periodic spline. Only used if sinusoid is FALSE.
intercept: If TRUE, a column of ones will be included in the basis.

Value

A matrix with a row for each sample and a column for each component of the time decomposition.

Examples

# create an example data frame
nSamples = 12
d = data.frame(
  sample = paste0('sample_', 1:nSamples),
  genotype = factor(rep(c('WT', 'KO'), each = nSamples / 2),
                    levels = c('WT', 'KO')),
  zt = rep(seq(0, 24 - 24 / nSamples * 2, 24 / nSamples * 2), times = 2),
  stringsAsFactors = FALSE)

# call limorhyde
limo = limorhyde(d$zt, 'zt_')
d = cbind(d, limo)

# create a design matrix that could be used with methods such as limma
design = model.matrix(~ genotype * (zt_cos + zt_sin), data = d)