viterbi: Calculate most probable sequence of genotypes

Description

Uses a hidden Markov model to calculate arg max Pr(g | O) where g is the underlying sequence of true genotypes and O is the observed multipoint marker data, with possible allowance for genotyping errors.

Usage

viterbi(
  cross,
  map = NULL,
  error_prob = 0.0001,
  map_function = c("haldane", "kosambi", "c-f", "morgan"),
  lowmem = FALSE,
  quiet = TRUE,
  cores = 1
)

Value

An object of class "viterbi": a list of two-dimensional arrays of imputed genotypes, individuals x positions. Also contains three attributes:

crosstype - The cross type of the input cross.
is_x_chr - Logical vector indicating whether chromosomes are to be treated as the X chromosome or not, from input cross.
alleles - Vector of allele codes, from input cross.

Arguments

cross: Object of class "cross2". For details, see the R/qtl2 developer guide.
map: Genetic map of markers. May include pseudomarker locations (that is, locations that are not within the marker genotype data). If NULL, the genetic map in cross is used.
error_prob: Assumed genotyping error probability
map_function: Character string indicating the map function to use to convert genetic distances to recombination fractions.
lowmem: If FALSE, split individuals into groups with common sex and crossinfo and then precalculate the transition matrices for a chromosome; potentially a lot faster but using more memory.
quiet: If FALSE, print progress messages.
cores: Number of CPU cores to use, for parallel calculations. (If 0, use parallel::detectCores().) Alternatively, this can be links to a set of cluster sockets, as produced by parallel::makeCluster().

Details

We use a hidden Markov model to find, for each individual on each chromosome, the most probable sequence of underlying genotypes given the observed marker data.

Note that we break ties at random, and our method for doing this may introduce some bias.

Consider the results with caution; the most probable sequence can have very low probability, and can have features that are quite unusual (for example, the number of recombination events can be too small). In most cases, the results of a single imputation with sim_geno() will be more realistic.

Examples

Run this code

grav2 <- read_cross2(system.file("extdata", "grav2.zip", package="qtl2"))
map_w_pmar <- insert_pseudomarkers(grav2$gmap, step=1)
g <- viterbi(grav2, map_w_pmar, error_prob=0.002)