Suggestions: Include a Chao-Jaccard abundance index function for repOverlap #85
Description
The formula for the Jaccard index (though it might technically be a coefficient) used is a comparison of sequence occurrence, regardless of how often it appears in the set (comparison of unique sequence use) to compare similarity.
However, the Chao Jaccard Index is an abundance based index which takes into account the number of times a particular combination occurs.
We frequently use both of these metrics to calculate repertoire overlap.
I think the Chao Jaccard Index can be included in the overlap.R file by adding the following R code:
Change line 98 of overlap.R to:
.method = c("public", "overlap", "jaccard", "chao_jaccard_abundance_index", "tversky", "cosine", "morisita", "inc+public", "inc+morisita"),
Add the following functions in at line 222 of overlap.R:
chao_jaccard_abundance_index <- function(.x, .y) {
UseMethod("chao_jaccard_abundance_index")
}
chao_jaccard_abundance_index.default <- function(.x, .y) {
.x <- collect(.x, n = Inf)
.y <- collect(.y, n = Inf)
intersection <- nrow(dplyr::intersect(.x, .y))
proportion_of_x_in_y_counting_all_seqs <- intersection / nrow(.y)
proportion_of_y_in_x_counting_all_seqs <- intersection / nrow(.x)
(proportion_of_x_in_y_counting_all_seqs * proportion_of_y_in_x_counting_all_seqs) / (proportion_of_x_in_y_counting_all_seqs + proportion_of_y_in_x_counting_all_seqs - (proportion_of_x_in_y_counting_all_seqs * proportion_of_y_in_x_counting_all_seqs))
}
chao_jaccard_abundance_index.character <- function(.x, .y) {
intersection <- nrow(dplyr::intersect(.x, .y))
proportion_of_x_in_y_counting_all_seqs <- intersection / nrow(.y)
proportion_of_y_in_x_counting_all_seqs <- intersection / nrow(.x)
(proportion_of_x_in_y_counting_all_seqs * proportion_of_y_in_x_counting_all_seqs) / (proportion_of_x_in_y_counting_all_seqs + proportion_of_y_in_x_counting_all_seqs - (proportion_of_x_in_y_counting_all_seqs * proportion_of_y_in_x_counting_all_seqs))
}
Clarifying Points
I'm happy to add a pull request to add this, but I'm not sure if this chao_jaccard_abundance_index function is correct. The math is right, but it requires taking the sum of all the sequences in X, Y not the clones. I'm not sure exactly what the function is taking in, but I think it is on the clonal level, which would not be right.