Model Selection - Expectation Maximization - Choose your distribution and type

This function was made to run a subset of models based on a selected distribution and type. There are many limitations to this function to make this tractable, as there are 128 models that could be run with our package. Here we do not include models or comparisons we found unhelpful, this includes the nQuire implementation and log-likelihood ratio tests.

Usage

bestquack(
  xm,
  distribution,
  type,
  uniform,
  mixtures = c("diploid", "triploid", "tetraploid", "hexaploid", "pentaploid"),
  samplename,
  trunc = c(0, 0),
  lowvar = FALSE,
  tau = NA,
  error = NA
)

Arguments

xm: Matrix with two columns with total coverage and coverage for a randomly sampled allele.
distribution: May be set to normal, beta, or beta-binomial. We do not include the implementation with nQuire.
type: May be equal to fixed, fixed_2, or fixed_3.
uniform: If equal to 1, a uniform mixture is included. If equal to 0, no uniform mixture is included.
mixtures: Defaults to c("diploid", "triploid", "tetraploid", "hexaploid", "pentaploid").
samplename: Name of sample to be included in output.
trunc: List of two values representing the lower and upper bounds for allele frequency truncation ,\(c_{L}\) and \(c_{U}\). If allele frequency truncation was done to remove error, then you do not need to truncate the expected. If no truncation has been done, this should be set to c(0,0), which is the default.
lowvar: Default to FALSE. When false, variance is equal to 0.01. If set to TRUE and tau and error are not provided, the variance will be set as 0.001.
tau: Sequencing overdispersion parameter. If tau and error are provided, the variance of each mixture will be inferred from these values. If not, the variance by default is equal to 0.01 or 0.001.
error: Sequencing error rate. If tau and error are provided, the variance of each mixture will be inferred from these values. If not, the variance by default is equal to 0.01 or 0.001.

Value

BIC scores and log-likelihood (LL) for the included mixture models. For BIC, the smallest score is the most likely model. For LL, the largest score is the most likely model.