
Filter individuals based on a relationship threshold
Source:R/filter_high_relatedness.R
filter_high_relatedness.RdThis function takes a matrix of x by y individuals containing relatedness coefficients and returns the maximum set of individuals that contains no relationships above the given threshold.
Arguments
- matrix
a square symmetric matrix of individuals containing relationship coefficients
- .x
a
gen_tibbleobject- kings_threshold
a threshold over which
- verbose
boolean whether to report to screen
Value
a list where '1' is individual ID's to retain, '2' is individual ID's to remove, and '3' is a boolean where individuals to keep are TRUE and individuals to remove are FALSE
Details
Individuals are removed greedily, as in
caret::findCorrelation(): individuals are considered in order of
decreasing mean relatedness, and for each pair related above
kings_threshold, the first individual is removed if its mean
relatedness is larger than the mean of the remaining matrix without the
second individual, and the second one otherwise. No pair above the
threshold is left, but the set of individuals kept is not guaranteed to
be the largest possible.
Missing relationship coefficients (e.g. KING kinship for pairs of individuals with no loci genotyped in both) are not evidence of relatedness: such pairs are treated as not related, with a warning.
Examples
example_gt <- load_example_gt("gen_tbl")
# Calculate relationship matrix
king_matrix <- example_gt %>% pairwise_king(as_matrix = TRUE)
# Filter individuals with threshold above 0.2
filter_high_relatedness(king_matrix, example_gt, kings_threshold = 0.2)
#> Warning: 2 pairs of individuals have missing kinship (e.g. no loci genotyped in both), so they could not be compared, and are not treated as related
#> [[1]]
#> [1] "d" "c" "g" "e" "b" "f"
#>
#> [[2]]
#> [1] "a"
#>
#> [[3]]
#> [1] FALSE TRUE TRUE TRUE TRUE TRUE TRUE
#>