Nothing
R/main.R
In idr2d: Irreproducible Discovery Rate for Genomic Interactions Data
Defines functions
print.idr2d_result_summary
summary.idr2d_result
estimate_idr
estimate_idr2d
estimate_idr1d
preprocess
establish_bijection
establish_bijection2d
establish_bijection1d
Documented in
establish_bijection
establish_bijection1d
establish_bijection2d
estimate_idr
estimate_idr1d
estimate_idr2d
preprocess
#' @title Finds One-to-One Correspondence between Peaks from Replicate 1
#' and 2
#'
#' @description
#' This method establishes a bijective assignment between peaks from
#' replicate 1 and 2. A peak in replicate 1 is assigned to a
#' peak in replicate 2 if and only if (1) they overlap (or the gap between the
#' peaks is less than or equal to \code{max_gap}), and (2) there is no other
#' peak in
#' replicate 2 that overlaps with the peak in replicate 1 and has a
#' lower \emph{ambiguity resolution value}.
#'
#' @inheritParams establish_overlap1d
#'
#' @return Data frames \code{rep1_df} and \code{rep2_df} with
#' the following columns:
#' \tabular{rll}{
#' column 1: \tab \code{chr} \tab character; genomic location of peak -
#' chromosome (e.g., \code{"chr3"})\cr
#' column 2: \tab \code{start} \tab integer; genomic location of peak -
#' start coordinate\cr
#' column 3: \tab \code{end} \tab integer; genomic location of peak -
#' end coordinate\cr
#' column 4: \tab \code{value} \tab numeric; p-value, FDR, or heuristic used
#' to rank the peaks\cr
#' column 5: \tab \code{rep_value} \tab numeric; value of corresponding
#' replicate peak. If no corresponding peak was found, \code{rep_value} is set
#' to \code{NA}.\cr
#' column 6: \tab \code{rank} \tab integer; rank of the peak, established by
#' value column, ascending order\cr
#' column 7: \tab \code{rep_rank} \tab integer; rank of corresponding
#' replicate peak. If no corresponding peak was found, \code{rep_rank} is
#' set to \code{NA}.\cr
#' column 8: \tab \code{idx} \tab integer; peak index, primary key\cr
#' column 9: \tab \code{rep_idx} \tab integer; specifies the index of the
#' corresponding peak in the other replicate (foreign key). If no
#' corresponding peak was found, \code{rep_idx} is set to \code{NA}.
#' }
#'
#' @examples
#' rep1_df <- idr2d:::chipseq$rep1_df
#' rep1_df$value <- preprocess(rep1_df$value, "log")
#'
#' rep2_df <- idr2d:::chipseq$rep2_df
#' rep2_df$value <- preprocess(rep2_df$value, "log")
#'
#' mapping <- establish_bijection1d(rep1_df, rep2_df)
#'
#' @export
establish_bijection1d <- function(rep1_df, rep2_df,
ambiguity_resolution_method = c("overlap",
"midpoint",
"value"),
max_gap = -1L) {
return(establish_bijection(rep1_df, rep2_df, "IDR1D",
ambiguity_resolution_method))
}
#' @title Finds One-to-One Correspondence between Interactions from Replicate 1
#' and 2
#'
#' @description
#' This method establishes a bijective assignment between interactions from
#' replicate 1 and 2. An interaction in replicate 1 is assigned to an
#' interaction in replicate 2 if and only if (1) both anchors of the
#' interactions
#' overlap (or the gap between anchor A/B in replicate 1 and 2 is less than
#' or equal to \code{max_gap}), and (2) there is no other interaction in
#' replicate 2 that overlaps with the interaction in replicate 1 and has a
#' lower \emph{ambiguity resolution value}.
#'
#' @inheritParams establish_overlap2d
#'
#' @return Data frames \code{rep1_df} and \code{rep2_df} with
#' the following columns:
#' \tabular{rll}{
#' column 1: \tab \code{chr_a} \tab character; genomic location of anchor A -
#' chromosome (e.g., \code{"chr3"})\cr
#' column 2: \tab \code{start_a} \tab integer; genomic location of anchor A -
#' start coordinate\cr
#' column 3: \tab \code{end_a} \tab integer; genomic location of anchor A -
#' end coordinate\cr
#' column 4: \tab \code{chr_b} \tab character; genomic location of anchor B -
#' chromosome (e.g., \code{"chr3"})\cr
#' column 5: \tab \code{start_b} \tab integer; genomic location of anchor B -
#' start coordinate\cr
#' column 6: \tab \code{end_b} \tab integer; genomic location of anchor B -
#' end coordinate\cr
#' column 7: \tab \code{value} \tab numeric; p-value, FDR, or heuristic used
#' to rank the interactions\cr
#' column 8: \tab \code{"rep_value"} \tab numeric; value of corresponding
#' replicate interaction. If no corresponding interaction was found,
#' \code{rep_value} is set to \code{NA}.\cr
#' column 9: \tab \code{"rank"} \tab integer; rank of the interaction,
#' established by value column, ascending order\cr
#' column 10: \tab \code{"rep_rank"} \tab integer; rank of corresponding
#' replicate interaction. If no corresponding interaction was found,
#' \code{rep_rank} is set to \code{NA}.\cr
#' column 11: \tab \code{"idx"} \tab integer; interaction index,
#' primary key\cr
#' column 12: \tab \code{"rep_idx"} \tab integer; specifies the index of the
#' corresponding interaction in the other replicate (foreign key). If no
#' corresponding interaction was found, \code{rep_idx} is set to \code{NA}.
#' }
#'
#' @examples
#' rep1_df <- idr2d:::chiapet$rep1_df
#' rep1_df$fdr <- preprocess(rep1_df$fdr, "log_additive_inverse")
#'
#' rep2_df <- idr2d:::chiapet$rep2_df
#' rep2_df$fdr <- preprocess(rep2_df$fdr, "log_additive_inverse")
#'
#' mapping <- establish_bijection2d(rep1_df, rep2_df)
#'
#' @export
establish_bijection2d <- function(rep1_df, rep2_df,
ambiguity_resolution_method = c("overlap",
"midpoint",
"value"),
max_gap = -1L) {
return(establish_bijection(rep1_df, rep2_df, "IDR2D",
ambiguity_resolution_method))
}
#' @title Finds One-to-One Correspondence between Peaks or interactions
#' from Replicate 1 and 2
#'
#' @description
#' This method establishes a bijective assignment between observations
#' (genomic peaks in case of ChIP-seq, genomic interactions in case of
#' ChIA-PET, HiChIP, and Hi-C) from
#' replicate 1 and 2. An observation in replicate 1 is assigned to an
#' observation in replicate 2 if and only if (1) the observation loci in both
#' replicates overlap (or the gap between them is less than
#' or equal to \code{max_gap}), and (2) there is no other observation in
#' replicate 2 that overlaps with the observation in replicate 1 and has a
#' lower \emph{ambiguity resolution value}.
#'
#'@param rep1_df data frame of observations (i.e., genomic peaks or genomic
#' interactions) of
#' replicate 1. If \code{analysis_type} is IDR1D, the columns of \code{rep1_df}
#' are described in \code{\link{establish_bijection1d}}, otherwise in
#' \code{\link{establish_bijection2d}}
#' @param rep2_df data frame of observations (i.e., genomic peaks or genomic
#' interactions) of replicate 2. Same columns as \code{rep1_df}.
#' @param analysis_type "IDR2D" for genomic interaction data sets,
#' "IDR1D" for genomic peak data sets
#' @param ambiguity_resolution_method defines how ambiguous assignments
#' (when one interaction or peak in replicate 1 overlaps with
#' multiple interactions or peaks in replicate 2 or vice versa)
#' are resolved. For available methods, see \code{\link{establish_overlap1d}} or
#' \code{\link{establish_overlap2d}}, respectively.
#'
#' @inheritParams determine_anchor_overlap
#'
#' @return See \code{\link{establish_bijection1d}} or
#' \code{\link{establish_bijection2d}}, respectively.
#'
#' @examples
#' rep1_df <- idr2d:::chipseq$rep1_df
#' rep1_df$value <- preprocess(rep1_df$value, "log")
#'
#' rep2_df <- idr2d:::chipseq$rep2_df
#' rep2_df$value <- preprocess(rep2_df$value, "log")
#'
#' mapping <- establish_bijection(rep1_df, rep2_df, analysis_type = "IDR1D")
#'
#' @importFrom dplyr arrange
#' @importFrom dplyr desc
#' @importFrom dplyr group_by
#' @importFrom dplyr slice
#' @importFrom dplyr select
#' @importFrom futile.logger flog.warn
#' @importFrom magrittr "%>%"
#' @export
establish_bijection <- function(rep1_df, rep2_df,
analysis_type = c("IDR1D", "IDR2D"),
ambiguity_resolution_method = c("overlap",
"midpoint",
"value"),
max_gap = -1L) {
# avoid CRAN warnings
rep1_idx <- rep2_idx <- arv <- value <- NULL
# argument handling
analysis_type <- match.arg(analysis_type, choices = c("IDR1D", "IDR2D"))
if (analysis_type == "IDR1D") {
columns <- c("chr", "start", "end", "value")
} else if (analysis_type == "IDR2D") {
columns <- c("chr_a", "start_a", "end_a",
"chr_b", "start_b", "end_b", "value")
}
if (nrow(rep1_df) > 0 && nrow(rep2_df) > 0) {
# remove irrelevant columns, rename relevant columns
rep1_df <- rep1_df[, seq_len(length(columns))]
rep2_df <- rep2_df[, seq_len(length(columns))]
colnames(rep1_df) <- columns
colnames(rep2_df) <- columns
# shuffle to break preexisting order
rep1_df <- rep1_df[sample.int(nrow(rep1_df)), ]
rep2_df <- rep2_df[sample.int(nrow(rep2_df)), ]
# sort by value column, descending order (higher values are better)
rep1_df <- dplyr::arrange(rep1_df, dplyr::desc(value))
rep2_df <- dplyr::arrange(rep2_df, dplyr::desc(value))
# add idx column
rep1_df$idx <- seq_len(nrow(rep1_df))
rep2_df$idx <- seq_len(nrow(rep2_df))
# add rank column
rep1_df$rank <- seq_len(nrow(rep1_df))
rep2_df$rank <- seq_len(nrow(rep2_df))
if (analysis_type == "IDR1D") {
pairs_df <- establish_overlap1d(rep1_df, rep2_df,
ambiguity_resolution_method,
max_gap)
} else if (analysis_type == "IDR2D") {
pairs_df <- establish_overlap2d(rep1_df, rep2_df,
ambiguity_resolution_method,
max_gap)
}
top_pairs_df <- pairs_df %>% dplyr::group_by(rep1_idx) %>%
dplyr::slice(which.min(arv))
top_pairs_df <- top_pairs_df %>% dplyr::group_by(rep2_idx) %>%
dplyr::slice(which.min(arv))
if (nrow(top_pairs_df) != length(unique(top_pairs_df$rep1_idx)) &&
nrow(top_pairs_df) != length(unique(top_pairs_df$rep2_idx))) {
futile.logger::flog.warn("ambiguous interaction assignments")
}
rep1_df$rep_idx <- as.integer(NA)
rep2_df$rep_idx <- as.integer(NA)
rep1_df$rep_idx[top_pairs_df$rep1_idx] <- top_pairs_df$rep2_idx
rep2_df$rep_idx[top_pairs_df$rep2_idx] <- top_pairs_df$rep1_idx
rep1_df$rep_rank <- as.integer(NA)
rep2_df$rep_rank <- as.integer(NA)
rep1_df$rep_rank[top_pairs_df$rep1_idx] <-
rep2_df$rank[top_pairs_df$rep2_idx]
rep2_df$rep_rank[top_pairs_df$rep2_idx] <-
rep1_df$rank[top_pairs_df$rep1_idx]
rep1_df$rep_value <- as.numeric(NA)
rep2_df$rep_value <- as.numeric(NA)
rep1_df$rep_value[top_pairs_df$rep1_idx] <-
rep2_df$value[top_pairs_df$rep2_idx]
rep2_df$rep_value[top_pairs_df$rep2_idx] <-
rep1_df$value[top_pairs_df$rep1_idx]
columns <- c(columns, "rep_value", "rank", "rep_rank", "idx", "rep_idx")
rep1_df <- dplyr::select(rep1_df, columns)
rep2_df <- dplyr::select(rep2_df, columns)
} else {
if (nrow(rep1_df) == 0) {
rep1_df$idx <- integer(0)
rep1_df$rep_idx <- integer(0)
rep1_df$rank <- integer(0)
rep1_df$rep_rank <- integer(0)
rep1_df$rep_value <- numeric(0)
} else {
# remove irrelevant columns, rename relevant columns
rep1_df <- rep1_df[, seq_len(length(columns))]
colnames(rep1_df) <- columns
columns <- c(columns, "rep_value", "rank",
"rep_rank", "idx", "rep_idx")
rep1_df$idx <- NA_integer_
rep1_df$rep_idx <- NA_integer_
rep1_df$rank <- NA_integer_
rep1_df$rep_rank <- NA_integer_
rep1_df$rep_value <- NA_real_
rep1_df <- dplyr::select(rep1_df, columns)
}
if (nrow(rep2_df) == 0) {
rep2_df$idx <- integer(0)
rep2_df$rep_idx <- integer(0)
rep2_df$rank <- integer(0)
rep2_df$rep_rank <- integer(0)
rep2_df$rep_value <- numeric(0)
} else {
# remove irrelevant columns, rename relevant columns
rep2_df <- rep2_df[, seq_len(length(columns))]
colnames(rep2_df) <- columns
columns <- c(columns, "rep_value", "rank",
"rep_rank", "idx", "rep_idx")
rep2_df$idx <- NA_integer_
rep2_df$rep_idx <- NA_integer_
rep2_df$rank <- NA_integer_
rep2_df$rep_rank <- NA_integer_
rep2_df$rep_value <- NA_real_
rep2_df <- dplyr::select(rep2_df, columns)
}
}
return(list(rep1_df = rep1_df, rep2_df = rep2_df))
}
#' @title Prepares Data for IDR Analysis
#'
#' @description
#' This method removes invalid values, establishes the correct
#' ranking, and breaks ties prior to IDR analysis.
#'
#' \code{Inf} and \code{-Inf} are replaced by \code{max(x) * max_factor}
#' and \code{min(x) / max_factor}, respectively.
#'
#' \code{NA} values in \code{x} are replaced by \code{mean(x)}.
#'
#' All values in \code{x} are transformed using the transformation specified
#' in \code{value_transformation}.
#'
#' Lastly, a small amount of noise is added to \code{x} to break ties. The
#' magnitude of the noise is controlled by \code{jitter_factor}.
#'
#' @param x numeric vector of values
#' @param value_transformation the values in \code{x} have to be transformed in
#' a way such that when ordered in descending order, more significant
#' interactions end up on top of the list. If the values in \code{x} are
#' p-values, \code{"log_additive_inverse"} is recommended. The following
#' transformations are supported:
#' \tabular{rl}{
#' \code{"identity"} \tab no transformation is performed on \code{x}\cr
#' \code{"additive_inverse"} \tab \code{x. = -x}\cr
#' \code{"multiplicative_inverse"} \tab \code{x. = 1 / x}\cr
#' \code{"log"} \tab \code{x. = log(x)}. Note: zeros are replaced by
#' \code{.Machine$double.xmin}\cr
#' \code{"log_additive_inverse"} \tab \code{x. = -log(x)}, recommended if
#' \code{x} are p-values. Note: zeros are replaced by
#' \code{.Machine$double.xmin}
#' }
#'
#' either \code{"ascending"} (more significant
#' interactions have lower value in \code{value} column) or \code{"descending"}
#' (more significant interactions have higher value in \code{value} column)
#' @param max_factor numeric; controls the replacement values for \code{Inf}
#' and \code{-Inf}. \code{Inf} are replaced by \code{max(x) * max_factor} and
#' \code{-Inf} are replaced by \code{min(x) / max_factor}.
#' @param jitter_factor numeric; controls the magnitude of the noise that
#' is added to \code{x}. This is done to break ties in \code{x}. Set
#' \code{jitter_factor = NULL} for no jitter.
#'
#' @return numeric vector; transformed and stripped values of \code{x}, ready
#' for IDR analysis
#'
#' @examples
#' rep1_df <- idr2d:::chiapet$rep1_df
#' rep1_df$fdr <- preprocess(rep1_df$fdr, "log_additive_inverse")
#'
#' @importFrom futile.logger flog.warn
#' @importFrom stringr str_trim
#' @export
preprocess <- function(x, value_transformation = c("identity",
"additive_inverse",
"multiplicative_inverse",
"log",
"log_additive_inverse"),
max_factor = 1.5,
jitter_factor = 0.0001) {
# argument handling
value_transformation <- match.arg(value_transformation,
choices = c("identity",
"additive_inverse",
"multiplicative_inverse",
"log",
"log_additive_inverse"))
# replace Inf with real values
max_value <- max(x[x != Inf], na.rm = TRUE)
x[x == Inf] <- max_value * max_factor
# replace -Inf with real values
min_value <- min(x[x != -Inf], na.rm = TRUE)
x[x == -Inf] <- min_value / max_factor
# replace NA with mean
x[is.na(x)] <- mean(x, na.rm = TRUE)
if (value_transformation == "additive_inverse") {
x <- (-1) * x
} else if (value_transformation == "multiplicative_inverse") {
x <- 1 / x
} else if (value_transformation == "log") {
x[x <= .Machine$double.xmin] <- .Machine$double.xmin
x <- log(x)
} else if (value_transformation == "log_additive_inverse") {
x[x <= .Machine$double.xmin] <- .Machine$double.xmin
x <- (-1) * log(x)
}
if (length(unique(x)) < 10) {
futile.logger::flog.warn("low complexity value distribution")
}
# add jitter to break ties
if (!is.null(jitter_factor)) {
invisible(tryCatch({
x <- jitter(x, factor = jitter_factor)
},
error = function(e) {
futile.logger::flog.warn(stringr::str_trim(e))
}))
}
return(x)
}
#' @title Estimates IDR for Genomic Peak Data
#'
#' @description
#' This method estimates Irreproducible Discovery Rates (IDR) for peaks in
#' replicated ChIP-seq experiments.
#'
#' @references
#' Q. Li, J. B. Brown, H. Huang and P. J. Bickel. (2011) Measuring
#' reproducibility of high-throughput experiments. Annals of Applied
#' Statistics, Vol. 5, No. 3, 1752-1779.
#'
#' @param remove_nonstandard_chromosomes removes peaks containing
#' genomic locations on non-standard chromosomes using
#' \code{\link[GenomeInfoDb:seqlevels-wrappers]{keepStandardChromosomes}}
#' (default is TRUE)
#' @param max_iteration integer; maximum number of iterations for
#' IDR estimation (defaults to 30)
#' @param local_idr see \code{\link[idr:est.IDR]{est.IDR}}
#'
#' @inheritParams establish_bijection1d
#' @inheritParams idr::est.IDR
#' @inheritParams preprocess
#'
#' @return List with three components, (\code{rep1_df}, \code{rep2_df},
#' and \code{analysis_type}) containing the interactions from input
#' data frames \code{rep1_df} and \code{rep2_df} with
#' the following additional columns:
#' \tabular{rll}{
#' column 1: \tab \code{chr} \tab character; genomic location of peak -
#' chromosome (e.g., \code{"chr3"})\cr
#' column 2: \tab \code{start} \tab integer; genomic location of peak -
#' start coordinate\cr
#' column 3: \tab \code{end} \tab integer; genomic location of peak -
#' end coordinate\cr
#' column 4: \tab \code{value} \tab numeric; p-value, FDR, or heuristic used
#' to rank the peaks\cr
#' column 5: \tab \code{rep_value} \tab numeric; value of corresponding
#' replicate peak. If no corresponding peak was found, \code{rep_value} is set
#' to \code{NA}.\cr
#' column 6: \tab \code{rank} \tab integer; rank of the peak, established by
#' value column, ascending order\cr
#' column 7: \tab \code{rep_rank} \tab integer; rank of corresponding
#' replicate peak. If no corresponding peak was found, \code{rep_rank} is
#' set to \code{NA}.\cr
#' column 8: \tab \code{idx} \tab integer; peak index, primary key\cr
#' column 9: \tab \code{rep_idx} \tab integer; specifies the index of the
#' corresponding peak in the other replicate (foreign key). If no
#' corresponding peak was found, \code{rep_idx} is set to \code{NA}.\cr
#' column 10: \tab \code{idr} \tab IDR of the peak and the
#' corresponding peak in the other replicate. If no corresponding
#' peak was found, \code{idr} is set to \code{NA}.
#' }
#'
#' @examples
#' idr_results <- estimate_idr1d(idr2d:::chipseq$rep1_df,
#' idr2d:::chipseq$rep2_df,
#' value_transformation = "log")
#' summary(idr_results)
#'
#' @export
estimate_idr1d <- function(rep1_df, rep2_df,
value_transformation = c("identity",
"additive_inverse",
"multiplicative_inverse",
"log",
"log_additive_inverse"),
ambiguity_resolution_method = c("overlap",
"midpoint",
"value"),
remove_nonstandard_chromosomes = TRUE,
max_factor = 1.5, jitter_factor = 0.0001,
max_gap = -1L,
mu = 0.1, sigma = 1.0, rho = 0.2, p = 0.5,
eps = 0.001, max_iteration = 30, local_idr = TRUE) {
return(estimate_idr(rep1_df, rep2_df, "IDR1D",
value_transformation,
ambiguity_resolution_method,
remove_nonstandard_chromosomes,
max_factor, jitter_factor, max_gap,
mu, sigma, rho, p,
eps, max_iteration, local_idr))
}
#' @title Estimates IDR for Genomic Interaction Data
#'
#' @description
#' This method estimates Irreproducible Discovery Rates (IDR) between
#' two replicates of experiments identifying genomic interactions, such as
#' Hi-C, ChIA-PET, and HiChIP.
#'
#' @references
#' Q. Li, J. B. Brown, H. Huang and P. J. Bickel. (2011) Measuring
#' reproducibility of high-throughput experiments. Annals of Applied
#' Statistics, Vol. 5, No. 3, 1752-1779.
#'
#' @param remove_nonstandard_chromosomes removes interactions
#' containing
#' genomic locations on non-standard chromosomes using
#' \code{\link[GenomeInfoDb:seqlevels-wrappers]{keepStandardChromosomes}}
#' (default is TRUE)
#' @param max_iteration integer; maximum number of iterations for
#' IDR estimation (defaults to 30)
#' @param local_idr see \code{\link[idr:est.IDR]{est.IDR}}
#'
#' @inheritParams establish_bijection2d
#' @inheritParams idr::est.IDR
#' @inheritParams preprocess
#'
#' @return List with three components, (\code{rep1_df}, \code{rep2_df},
#' and \code{analysis_type}) containing the interactions from input
#' data frames \code{rep1_df} and \code{rep2_df} with
#' the following additional columns:
#' \tabular{rll}{
#' column 1: \tab \code{chr_a} \tab character; genomic location of anchor A -
#' chromosome (e.g., \code{"chr3"})\cr
#' column 2: \tab \code{start_a} \tab integer; genomic location of anchor A -
#' start coordinate\cr
#' column 3: \tab \code{end_a} \tab integer; genomic location of anchor A -
#' end coordinate\cr
#' column 4: \tab \code{chr_b} \tab character; genomic location of anchor B -
#' chromosome (e.g., \code{"chr3"})\cr
#' column 5: \tab \code{start_b} \tab integer; genomic location of anchor B -
#' start coordinate\cr
#' column 6: \tab \code{end_b} \tab integer; genomic location of anchor B -
#' end coordinate\cr
#' column 7: \tab \code{value} \tab numeric; p-value, FDR, or heuristic used
#' to rank the interactions\cr
#' column 8: \tab \code{"rep_value"} \tab numeric; value of corresponding
#' replicate interaction. If no corresponding interaction was found,
#' \code{rep_value} is set to \code{NA}.\cr
#' column 9: \tab \code{"rank"} \tab integer; rank of the interaction,
#' established by value column, ascending order\cr
#' column 10: \tab \code{"rep_rank"} \tab integer; rank of corresponding
#' replicate interaction. If no corresponding interaction was found,
#' \code{rep_rank} is set to \code{NA}.\cr
#' column 11: \tab \code{"idx"} \tab integer; interaction index,
#' primary key\cr
#' column 12: \tab \code{"rep_idx"} \tab integer; specifies the index of the
#' corresponding interaction in the other replicate (foreign key). If no
#' corresponding interaction was found, \code{rep_idx} is set to \code{NA}.\cr
#' \code{idr} \tab IDR of the interaction and the
#' corresponding interaction in the other replicate. If no corresponding
#' interaction was found, \code{idr} is set to \code{NA}.
#' }
#'
#' @examples
#' idr_results <- estimate_idr2d(idr2d:::chiapet$rep1_df,
#' idr2d:::chiapet$rep2_df,
#' value_transformation = "log_additive_inverse")
#' summary(idr_results)
#'
#' @export
estimate_idr2d <- function(rep1_df, rep2_df,
value_transformation = c("identity",
"additive_inverse",
"multiplicative_inverse",
"log",
"log_additive_inverse"),
ambiguity_resolution_method = c("overlap",
"midpoint",
"value"),
remove_nonstandard_chromosomes = TRUE,
max_factor = 1.5, jitter_factor = 0.0001,
max_gap = -1L,
mu = 0.1, sigma = 1.0, rho = 0.2, p = 0.5,
eps = 0.001, max_iteration = 30, local_idr = TRUE) {
return(estimate_idr(rep1_df, rep2_df, "IDR2D",
value_transformation,
ambiguity_resolution_method,
remove_nonstandard_chromosomes,
max_factor, jitter_factor,
max_gap,
mu, sigma, rho, p,
eps, max_iteration, local_idr))
}
#' @title Estimates IDR for Genomic Peaks or Genomic Interactions
#' @references
#' Q. Li, J. B. Brown, H. Huang and P. J. Bickel. (2011) Measuring
#' reproducibility of high-throughput experiments. Annals of Applied
#' Statistics, Vol. 5, No. 3, 1752-1779.
#'
#' @param remove_nonstandard_chromosomes removes peaks and interactions
#' containing
#' genomic locations on non-standard chromosomes using
#' \code{\link[GenomeInfoDb:seqlevels-wrappers]{keepStandardChromosomes}}
#' (default is TRUE)
#' @param max_iteration integer; maximum number of iterations for
#' IDR estimation (defaults to 30)
#' @param local_idr see \code{\link[idr:est.IDR]{est.IDR}}
#'
#' @inheritParams idr::est.IDR
#' @inheritParams preprocess
#' @inheritParams establish_bijection
#'
#'
#' @return See \code{\link{estimate_idr1d}} or
#' \code{\link{estimate_idr2d}}, respectively.
#'
#' @examples
#' idr_results <- estimate_idr(idr2d:::chiapet$rep1_df,
#' idr2d:::chiapet$rep2_df,
#' analysis_type = "IDR2D",
#' value_transformation = "log_additive_inverse")
#' summary(idr_results)
#'
#' @importFrom dplyr arrange
#' @importFrom dplyr select
#' @importFrom dplyr filter
#' @importFrom futile.logger flog.warn
#' @importFrom stringr str_trim
#' @importFrom idr est.IDR
#' @export
estimate_idr <- function(rep1_df, rep2_df, analysis_type = "IDR2D",
value_transformation = c("identity",
"additive_inverse",
"multiplicative_inverse",
"log",
"log_additive_inverse"),
ambiguity_resolution_method = c("overlap",
"midpoint",
"value"),
remove_nonstandard_chromosomes = TRUE,
max_factor = 1.5, jitter_factor = 0.0001,
max_gap = -1L,
mu = 0.1, sigma = 1.0, rho = 0.2, p = 0.5,
eps = 0.001, max_iteration = 30, local_idr = TRUE) {
# avoid CRAN warnings
rep2_idx <- rep1_value <- rep2_value <- idr <- NULL
# argument handling
analysis_type <- match.arg(analysis_type, choices = c("IDR1D", "IDR2D"))
if (analysis_type == "IDR1D") {
columns <- c("chr", "start", "end", "value")
} else if (analysis_type == "IDR2D") {
columns <- c("chr_a", "start_a", "end_a",
"chr_b", "start_b", "end_b", "value")
}
if (nrow(rep1_df) > 0 && nrow(rep2_df) > 0) {
# remove irrelevant columns, rename relevant columns
rep1_df <- rep1_df[, seq_len(length(columns))]
rep2_df <- rep2_df[, seq_len(length(columns))]
colnames(rep1_df) <- columns
colnames(rep2_df) <- columns
if (remove_nonstandard_chromosomes) {
if (analysis_type == "IDR1D") {
rep1_df <- remove_nonstandard_chromosomes1d(rep1_df)
rep2_df <- remove_nonstandard_chromosomes1d(rep2_df)
} else if (analysis_type == "IDR2D") {
rep1_df <- remove_nonstandard_chromosomes2d(rep1_df)
rep2_df <- remove_nonstandard_chromosomes2d(rep2_df)
}
}
rep1_df$value <- preprocess(rep1_df$value,
value_transformation,
max_factor = max_factor,
jitter_factor = jitter_factor)
rep2_df$value <- preprocess(rep2_df$value,
value_transformation,
max_factor = max_factor,
jitter_factor = jitter_factor)
mapping <- establish_bijection(rep1_df, rep2_df, analysis_type,
ambiguity_resolution_method,
max_gap = max_gap)
if (nrow(mapping$rep1_df) > 0 && nrow(mapping$rep2_df) > 0) {
rep1_df <- mapping$rep1_df
rep2_df <- mapping$rep2_df
idx_df <- data.frame(
rep1_idx = rep1_df$idx,
rep2_idx = rep1_df$rep_idx,
rep1_value = rep1_df$value,
rep2_value = rep1_df$rep_value
)
idx_df <- dplyr::filter(idx_df,
!is.na(rep2_idx) & !is.infinite(rep2_idx))
if (nrow(idx_df) > 0) {
idr_matrix <- as.matrix(dplyr::select(idx_df,
rep1_value,
rep2_value))
if (length(unique(idr_matrix[, 1])) < 10 ||
length(unique(idr_matrix[, 2])) < 10) {
futile.logger::flog.warn("low complexity data set")
}
invisible(tryCatch({
idr_results <- idr::est.IDR(idr_matrix, mu, sigma, rho, p,
eps = eps,
max.ite = max_iteration)
if (local_idr) {
idx_df$idr <- idr_results$idr
} else {
idx_df$idr <- idr_results$IDR
}
},
error = function(e) {
idx_df$idr <- as.numeric(NA)
futile.logger::flog.warn(stringr::str_trim(e))
}))
} else {
idx_df$idr <- numeric(0)
}
if (nrow(rep1_df) > 0) {
rep1_df$idr <- as.numeric(NA)
if (nrow(idx_df) > 0) {
rep1_df$idr[idx_df$rep1_idx] <- idx_df$idr
rep1_df <- dplyr::arrange(rep1_df, idr)
}
} else {
rep1_df$idr <- numeric(0)
}
if (nrow(rep2_df) > 0) {
rep2_df$idr <- as.numeric(NA)
if (length(idx_df$idr) > 0) {
rep2_df$idr[idx_df$rep2_idx] <- idx_df$idr
rep2_df <- dplyr::arrange(rep2_df, idr)
}
} else {
rep2_df$idr <- numeric(0)
}
} else {
if (nrow(rep1_df) > 0) {
rep1_df$idr <- as.numeric(NA)
} else {
rep1_df$idr <- numeric(0)
}
if (nrow(rep2_df) > 0) {
rep2_df$idr <- as.numeric(NA)
} else {
rep2_df$idr <- numeric(0)
}
}
columns <- c(columns, "rep_value", "rank", "rep_rank",
"idx", "rep_idx", "idr")
if (analysis_type == "IDR1D") {
rep1_df <- dplyr::select(rep1_df, columns)
rep2_df <- dplyr::select(rep2_df, columns)
} else if (analysis_type == "IDR2D") {
rep1_df <- dplyr::select(rep1_df, columns)
rep2_df <- dplyr::select(rep2_df, columns)
}
} else {
if (nrow(rep1_df) == 0) {
rep1_df$idx <- integer(0)
rep1_df$rep_idx <- integer(0)
rep1_df$rank <- integer(0)
rep1_df$rep_rank <- integer(0)
rep1_df$rep_value <- numeric(0)
rep1_df$idr <- numeric(0)
} else {
# remove irrelevant columns, rename relevant columns
rep1_df <- rep1_df[, seq_len(length(columns))]
colnames(rep1_df) <- columns
columns <- c(columns, "rep_value", "rank",
"rep_rank", "idx", "rep_idx", "idr")
rep1_df$idx <- NA_integer_
rep1_df$rep_idx <- NA_integer_
rep1_df$rank <- NA_integer_
rep1_df$rep_rank <- NA_integer_
rep1_df$rep_value <- NA_real_
rep1_df$idr <- NA_real_
rep1_df <- dplyr::select(rep1_df, columns)
}
if (nrow(rep2_df) == 0) {
rep2_df$idx <- integer(0)
rep2_df$rep_idx <- integer(0)
rep2_df$rank <- integer(0)
rep2_df$rep_rank <- integer(0)
rep2_df$rep_value <- numeric(0)
rep2_df$idr <- numeric(0)
} else {
# remove irrelevant columns, rename relevant columns
rep2_df <- rep2_df[, seq_len(length(columns))]
colnames(rep2_df) <- columns
columns <- c(columns, "rep_value", "rank",
"rep_rank", "idx", "rep_idx", "idr")
rep2_df$idx <- NA_integer_
rep2_df$rep_idx <- NA_integer_
rep2_df$rank <- NA_integer_
rep2_df$rep_rank <- NA_integer_
rep2_df$rep_value <- NA_real_
rep2_df$idr <- NA_real_
rep2_df <- dplyr::select(rep2_df, columns)
}
}
res <- list(rep1_df = rep1_df, rep2_df = rep2_df,
analysis_type = analysis_type)
class(res) <- c("idr2d_result", class(res))
return(res)
}
#' @importFrom methods is
#' @importFrom dplyr filter
#' @export
summary.idr2d_result <- function(object, ...) {
idr <- NULL
stopifnot(methods::is(object, "idr2d_result"))
num_rep_df <- dplyr::filter(object$rep1_df, !is.na(idr))
num_sig_df <- dplyr::filter(num_rep_df, idr < 0.05)
num_high_sig_df <- dplyr::filter(num_rep_df, idr < 0.01)
res <- list(
analysis_type = object$analysis_type,
rep1_num_interactions = nrow(object$rep1_df),
rep2_num_interactions = nrow(object$rep2_df),
num_reproducible_interactions = nrow(num_rep_df),
num_significant_interactions = nrow(num_sig_df),
num_highly_significant_interactions = nrow(num_high_sig_df)
)
class(res) <- c("idr2d_result_summary", class(res))
return(res)
}
#' @importFrom methods is
#' @export
print.idr2d_result_summary <- function(x, ...) {
stopifnot(methods::is(x, "idr2d_result_summary"))
cat("analysis type: ", x$analysis_type, "\n", sep = "")
if (x$analysis_type == "IDR2D Hi-C") {
cat("number of blocks: ",
x$num_blocks, "\n", sep = "")
cat("number of blocks with significant IDR (IDR < 0.05): ",
x$num_significant_blocks, "\n", sep = "")
cat("number of blocks with highly significant IDR (IDR < 0.01): ",
x$num_highly_significant_blocks, "\n", sep = "")
cat("percentage of blocks with significant IDR (IDR < 0.05): ",
round(x$num_significant_blocks / x$num_blocks * 100, digits = 2),
" %\n", sep = "")
} else {
cat("number of interactions in replicate 1: ",
x$rep1_num_interactions, "\n", sep = "")
cat("number of interactions in replicate 2: ",
x$rep2_num_interactions, "\n", sep = "")
cat("number of reproducible interactions: ",
x$num_reproducible_interactions, "\n", sep = "")
cat("number of interactions with significant IDR (IDR < 0.05): ",
x$num_significant_interactions, "\n", sep = "")
cat("number of interactions with highly significant IDR (IDR < 0.01): ",
x$num_highly_significant_interactions, "\n", sep = "")
cat("percentage of interactions with significant IDR (IDR < 0.05): ",
round(x$num_significant_interactions /
max(x$rep1_num_interactions, x$rep2_num_interactions) * 100,
digits = 2),
" %\n", sep = "")
}
}
Try the idr2d package in your browser
Any scripts or data that you put into this service are public.
idr2d documentation built on Nov. 8, 2020, 6:16 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Defines functions print.idr2d_result_summary summary.idr2d_result estimate_idr estimate_idr2d estimate_idr1d preprocess establish_bijection establish_bijection2d establish_bijection1d
Documented in establish_bijection establish_bijection1d establish_bijection2d estimate_idr estimate_idr1d estimate_idr2d preprocess
#' @title Finds One-to-One Correspondence between Peaks from Replicate 1
#' and 2
#'
#' @description
#' This method establishes a bijective assignment between peaks from
#' replicate 1 and 2. A peak in replicate 1 is assigned to a
#' peak in replicate 2 if and only if (1) they overlap (or the gap between the
#' peaks is less than or equal to \code{max_gap}), and (2) there is no other
#' peak in
#' replicate 2 that overlaps with the peak in replicate 1 and has a
#' lower \emph{ambiguity resolution value}.
#'
#' @inheritParams establish_overlap1d
#'
#' @return Data frames \code{rep1_df} and \code{rep2_df} with
#' the following columns:
#' \tabular{rll}{
#' column 1: \tab \code{chr} \tab character; genomic location of peak -
#' chromosome (e.g., \code{"chr3"})\cr
#' column 2: \tab \code{start} \tab integer; genomic location of peak -
#' start coordinate\cr
#' column 3: \tab \code{end} \tab integer; genomic location of peak -
#' end coordinate\cr
#' column 4: \tab \code{value} \tab numeric; p-value, FDR, or heuristic used
#' to rank the peaks\cr
#' column 5: \tab \code{rep_value} \tab numeric; value of corresponding
#' replicate peak. If no corresponding peak was found, \code{rep_value} is set
#' to \code{NA}.\cr
#' column 6: \tab \code{rank} \tab integer; rank of the peak, established by
#' value column, ascending order\cr
#' column 7: \tab \code{rep_rank} \tab integer; rank of corresponding
#' replicate peak. If no corresponding peak was found, \code{rep_rank} is
#' set to \code{NA}.\cr
#' column 8: \tab \code{idx} \tab integer; peak index, primary key\cr
#' column 9: \tab \code{rep_idx} \tab integer; specifies the index of the
#' corresponding peak in the other replicate (foreign key). If no
#' corresponding peak was found, \code{rep_idx} is set to \code{NA}.
#' }
#'
#' @examples
#' rep1_df <- idr2d:::chipseq$rep1_df
#' rep1_df$value <- preprocess(rep1_df$value, "log")
#'
#' rep2_df <- idr2d:::chipseq$rep2_df
#' rep2_df$value <- preprocess(rep2_df$value, "log")
#'
#' mapping <- establish_bijection1d(rep1_df, rep2_df)
#'
#' @export
establish_bijection1d <- function(rep1_df, rep2_df,
ambiguity_resolution_method = c("overlap",
"midpoint",
"value"),
max_gap = -1L) {
return(establish_bijection(rep1_df, rep2_df, "IDR1D",
ambiguity_resolution_method))
}
#' @title Finds One-to-One Correspondence between Interactions from Replicate 1
#' and 2
#'
#' @description
#' This method establishes a bijective assignment between interactions from
#' replicate 1 and 2. An interaction in replicate 1 is assigned to an
#' interaction in replicate 2 if and only if (1) both anchors of the
#' interactions
#' overlap (or the gap between anchor A/B in replicate 1 and 2 is less than
#' or equal to \code{max_gap}), and (2) there is no other interaction in
#' replicate 2 that overlaps with the interaction in replicate 1 and has a
#' lower \emph{ambiguity resolution value}.
#'
#' @inheritParams establish_overlap2d
#'
#' @return Data frames \code{rep1_df} and \code{rep2_df} with
#' the following columns:
#' \tabular{rll}{
#' column 1: \tab \code{chr_a} \tab character; genomic location of anchor A -
#' chromosome (e.g., \code{"chr3"})\cr
#' column 2: \tab \code{start_a} \tab integer; genomic location of anchor A -
#' start coordinate\cr
#' column 3: \tab \code{end_a} \tab integer; genomic location of anchor A -
#' end coordinate\cr
#' column 4: \tab \code{chr_b} \tab character; genomic location of anchor B -
#' chromosome (e.g., \code{"chr3"})\cr
#' column 5: \tab \code{start_b} \tab integer; genomic location of anchor B -
#' start coordinate\cr
#' column 6: \tab \code{end_b} \tab integer; genomic location of anchor B -
#' end coordinate\cr
#' column 7: \tab \code{value} \tab numeric; p-value, FDR, or heuristic used
#' to rank the interactions\cr
#' column 8: \tab \code{"rep_value"} \tab numeric; value of corresponding
#' replicate interaction. If no corresponding interaction was found,
#' \code{rep_value} is set to \code{NA}.\cr
#' column 9: \tab \code{"rank"} \tab integer; rank of the interaction,
#' established by value column, ascending order\cr
#' column 10: \tab \code{"rep_rank"} \tab integer; rank of corresponding
#' replicate interaction. If no corresponding interaction was found,
#' \code{rep_rank} is set to \code{NA}.\cr
#' column 11: \tab \code{"idx"} \tab integer; interaction index,
#' primary key\cr
#' column 12: \tab \code{"rep_idx"} \tab integer; specifies the index of the
#' corresponding interaction in the other replicate (foreign key). If no
#' corresponding interaction was found, \code{rep_idx} is set to \code{NA}.
#' }
#'
#' @examples
#' rep1_df <- idr2d:::chiapet$rep1_df
#' rep1_df$fdr <- preprocess(rep1_df$fdr, "log_additive_inverse")
#'
#' rep2_df <- idr2d:::chiapet$rep2_df
#' rep2_df$fdr <- preprocess(rep2_df$fdr, "log_additive_inverse")
#'
#' mapping <- establish_bijection2d(rep1_df, rep2_df)
#'
#' @export
establish_bijection2d <- function(rep1_df, rep2_df,
ambiguity_resolution_method = c("overlap",
"midpoint",
"value"),
max_gap = -1L) {
return(establish_bijection(rep1_df, rep2_df, "IDR2D",
ambiguity_resolution_method))
}
#' @title Finds One-to-One Correspondence between Peaks or interactions
#' from Replicate 1 and 2
#'
#' @description
#' This method establishes a bijective assignment between observations
#' (genomic peaks in case of ChIP-seq, genomic interactions in case of
#' ChIA-PET, HiChIP, and Hi-C) from
#' replicate 1 and 2. An observation in replicate 1 is assigned to an
#' observation in replicate 2 if and only if (1) the observation loci in both
#' replicates overlap (or the gap between them is less than
#' or equal to \code{max_gap}), and (2) there is no other observation in
#' replicate 2 that overlaps with the observation in replicate 1 and has a
#' lower \emph{ambiguity resolution value}.
#'
#'@param rep1_df data frame of observations (i.e., genomic peaks or genomic
#' interactions) of
#' replicate 1. If \code{analysis_type} is IDR1D, the columns of \code{rep1_df}
#' are described in \code{\link{establish_bijection1d}}, otherwise in
#' \code{\link{establish_bijection2d}}
#' @param rep2_df data frame of observations (i.e., genomic peaks or genomic
#' interactions) of replicate 2. Same columns as \code{rep1_df}.
#' @param analysis_type "IDR2D" for genomic interaction data sets,
#' "IDR1D" for genomic peak data sets
#' @param ambiguity_resolution_method defines how ambiguous assignments
#' (when one interaction or peak in replicate 1 overlaps with
#' multiple interactions or peaks in replicate 2 or vice versa)
#' are resolved. For available methods, see \code{\link{establish_overlap1d}} or
#' \code{\link{establish_overlap2d}}, respectively.
#'
#' @inheritParams determine_anchor_overlap
#'
#' @return See \code{\link{establish_bijection1d}} or
#' \code{\link{establish_bijection2d}}, respectively.
#'
#' @examples
#' rep1_df <- idr2d:::chipseq$rep1_df
#' rep1_df$value <- preprocess(rep1_df$value, "log")
#'
#' rep2_df <- idr2d:::chipseq$rep2_df
#' rep2_df$value <- preprocess(rep2_df$value, "log")
#'
#' mapping <- establish_bijection(rep1_df, rep2_df, analysis_type = "IDR1D")
#'
#' @importFrom dplyr arrange
#' @importFrom dplyr desc
#' @importFrom dplyr group_by
#' @importFrom dplyr slice
#' @importFrom dplyr select
#' @importFrom futile.logger flog.warn
#' @importFrom magrittr "%>%"
#' @export
establish_bijection <- function(rep1_df, rep2_df,
analysis_type = c("IDR1D", "IDR2D"),
ambiguity_resolution_method = c("overlap",
"midpoint",
"value"),
max_gap = -1L) {
# avoid CRAN warnings
rep1_idx <- rep2_idx <- arv <- value <- NULL
# argument handling
analysis_type <- match.arg(analysis_type, choices = c("IDR1D", "IDR2D"))
if (analysis_type == "IDR1D") {
columns <- c("chr", "start", "end", "value")
} else if (analysis_type == "IDR2D") {
columns <- c("chr_a", "start_a", "end_a",
"chr_b", "start_b", "end_b", "value")
}
if (nrow(rep1_df) > 0 && nrow(rep2_df) > 0) {
# remove irrelevant columns, rename relevant columns
rep1_df <- rep1_df[, seq_len(length(columns))]
rep2_df <- rep2_df[, seq_len(length(columns))]
colnames(rep1_df) <- columns
colnames(rep2_df) <- columns
# shuffle to break preexisting order
rep1_df <- rep1_df[sample.int(nrow(rep1_df)), ]
rep2_df <- rep2_df[sample.int(nrow(rep2_df)), ]
# sort by value column, descending order (higher values are better)
rep1_df <- dplyr::arrange(rep1_df, dplyr::desc(value))
rep2_df <- dplyr::arrange(rep2_df, dplyr::desc(value))
# add idx column
rep1_df$idx <- seq_len(nrow(rep1_df))
rep2_df$idx <- seq_len(nrow(rep2_df))
# add rank column
rep1_df$rank <- seq_len(nrow(rep1_df))
rep2_df$rank <- seq_len(nrow(rep2_df))
if (analysis_type == "IDR1D") {
pairs_df <- establish_overlap1d(rep1_df, rep2_df,
ambiguity_resolution_method,
max_gap)
} else if (analysis_type == "IDR2D") {
pairs_df <- establish_overlap2d(rep1_df, rep2_df,
ambiguity_resolution_method,
max_gap)
}
top_pairs_df <- pairs_df %>% dplyr::group_by(rep1_idx) %>%
dplyr::slice(which.min(arv))
top_pairs_df <- top_pairs_df %>% dplyr::group_by(rep2_idx) %>%
dplyr::slice(which.min(arv))
if (nrow(top_pairs_df) != length(unique(top_pairs_df$rep1_idx)) &&
nrow(top_pairs_df) != length(unique(top_pairs_df$rep2_idx))) {
futile.logger::flog.warn("ambiguous interaction assignments")
}
rep1_df$rep_idx <- as.integer(NA)
rep2_df$rep_idx <- as.integer(NA)
rep1_df$rep_idx[top_pairs_df$rep1_idx] <- top_pairs_df$rep2_idx
rep2_df$rep_idx[top_pairs_df$rep2_idx] <- top_pairs_df$rep1_idx
rep1_df$rep_rank <- as.integer(NA)
rep2_df$rep_rank <- as.integer(NA)
rep1_df$rep_rank[top_pairs_df$rep1_idx] <-
rep2_df$rank[top_pairs_df$rep2_idx]
rep2_df$rep_rank[top_pairs_df$rep2_idx] <-
rep1_df$rank[top_pairs_df$rep1_idx]
rep1_df$rep_value <- as.numeric(NA)
rep2_df$rep_value <- as.numeric(NA)
rep1_df$rep_value[top_pairs_df$rep1_idx] <-
rep2_df$value[top_pairs_df$rep2_idx]
rep2_df$rep_value[top_pairs_df$rep2_idx] <-
rep1_df$value[top_pairs_df$rep1_idx]
columns <- c(columns, "rep_value", "rank", "rep_rank", "idx", "rep_idx")
rep1_df <- dplyr::select(rep1_df, columns)
rep2_df <- dplyr::select(rep2_df, columns)
} else {
if (nrow(rep1_df) == 0) {
rep1_df$idx <- integer(0)
rep1_df$rep_idx <- integer(0)
rep1_df$rank <- integer(0)
rep1_df$rep_rank <- integer(0)
rep1_df$rep_value <- numeric(0)
} else {
# remove irrelevant columns, rename relevant columns
rep1_df <- rep1_df[, seq_len(length(columns))]
colnames(rep1_df) <- columns
columns <- c(columns, "rep_value", "rank",
"rep_rank", "idx", "rep_idx")
rep1_df$idx <- NA_integer_
rep1_df$rep_idx <- NA_integer_
rep1_df$rank <- NA_integer_
rep1_df$rep_rank <- NA_integer_
rep1_df$rep_value <- NA_real_
rep1_df <- dplyr::select(rep1_df, columns)
}
if (nrow(rep2_df) == 0) {
rep2_df$idx <- integer(0)
rep2_df$rep_idx <- integer(0)
rep2_df$rank <- integer(0)
rep2_df$rep_rank <- integer(0)
rep2_df$rep_value <- numeric(0)
} else {
# remove irrelevant columns, rename relevant columns
rep2_df <- rep2_df[, seq_len(length(columns))]
colnames(rep2_df) <- columns
columns <- c(columns, "rep_value", "rank",
"rep_rank", "idx", "rep_idx")
rep2_df$idx <- NA_integer_
rep2_df$rep_idx <- NA_integer_
rep2_df$rank <- NA_integer_
rep2_df$rep_rank <- NA_integer_
rep2_df$rep_value <- NA_real_
rep2_df <- dplyr::select(rep2_df, columns)
}
}
return(list(rep1_df = rep1_df, rep2_df = rep2_df))
}
#' @title Prepares Data for IDR Analysis
#'
#' @description
#' This method removes invalid values, establishes the correct
#' ranking, and breaks ties prior to IDR analysis.
#'
#' \code{Inf} and \code{-Inf} are replaced by \code{max(x) * max_factor}
#' and \code{min(x) / max_factor}, respectively.
#'
#' \code{NA} values in \code{x} are replaced by \code{mean(x)}.
#'
#' All values in \code{x} are transformed using the transformation specified
#' in \code{value_transformation}.
#'
#' Lastly, a small amount of noise is added to \code{x} to break ties. The
#' magnitude of the noise is controlled by \code{jitter_factor}.
#'
#' @param x numeric vector of values
#' @param value_transformation the values in \code{x} have to be transformed in
#' a way such that when ordered in descending order, more significant
#' interactions end up on top of the list. If the values in \code{x} are
#' p-values, \code{"log_additive_inverse"} is recommended. The following
#' transformations are supported:
#' \tabular{rl}{
#' \code{"identity"} \tab no transformation is performed on \code{x}\cr
#' \code{"additive_inverse"} \tab \code{x. = -x}\cr
#' \code{"multiplicative_inverse"} \tab \code{x. = 1 / x}\cr
#' \code{"log"} \tab \code{x. = log(x)}. Note: zeros are replaced by
#' \code{.Machine$double.xmin}\cr
#' \code{"log_additive_inverse"} \tab \code{x. = -log(x)}, recommended if
#' \code{x} are p-values. Note: zeros are replaced by
#' \code{.Machine$double.xmin}
#' }
#'
#' either \code{"ascending"} (more significant
#' interactions have lower value in \code{value} column) or \code{"descending"}
#' (more significant interactions have higher value in \code{value} column)
#' @param max_factor numeric; controls the replacement values for \code{Inf}
#' and \code{-Inf}. \code{Inf} are replaced by \code{max(x) * max_factor} and
#' \code{-Inf} are replaced by \code{min(x) / max_factor}.
#' @param jitter_factor numeric; controls the magnitude of the noise that
#' is added to \code{x}. This is done to break ties in \code{x}. Set
#' \code{jitter_factor = NULL} for no jitter.
#'
#' @return numeric vector; transformed and stripped values of \code{x}, ready
#' for IDR analysis
#'
#' @examples
#' rep1_df <- idr2d:::chiapet$rep1_df
#' rep1_df$fdr <- preprocess(rep1_df$fdr, "log_additive_inverse")
#'
#' @importFrom futile.logger flog.warn
#' @importFrom stringr str_trim
#' @export
preprocess <- function(x, value_transformation = c("identity",
"additive_inverse",
"multiplicative_inverse",
"log",
"log_additive_inverse"),
max_factor = 1.5,
jitter_factor = 0.0001) {
# argument handling
value_transformation <- match.arg(value_transformation,
choices = c("identity",
"additive_inverse",
"multiplicative_inverse",
"log",
"log_additive_inverse"))
# replace Inf with real values
max_value <- max(x[x != Inf], na.rm = TRUE)
x[x == Inf] <- max_value * max_factor
# replace -Inf with real values
min_value <- min(x[x != -Inf], na.rm = TRUE)
x[x == -Inf] <- min_value / max_factor
# replace NA with mean
x[is.na(x)] <- mean(x, na.rm = TRUE)
if (value_transformation == "additive_inverse") {
x <- (-1) * x
} else if (value_transformation == "multiplicative_inverse") {
x <- 1 / x
} else if (value_transformation == "log") {
x[x <= .Machine$double.xmin] <- .Machine$double.xmin
x <- log(x)
} else if (value_transformation == "log_additive_inverse") {
x[x <= .Machine$double.xmin] <- .Machine$double.xmin
x <- (-1) * log(x)
}
if (length(unique(x)) < 10) {
futile.logger::flog.warn("low complexity value distribution")
}
# add jitter to break ties
if (!is.null(jitter_factor)) {
invisible(tryCatch({
x <- jitter(x, factor = jitter_factor)
},
error = function(e) {
futile.logger::flog.warn(stringr::str_trim(e))
}))
}
return(x)
}
#' @title Estimates IDR for Genomic Peak Data
#'
#' @description
#' This method estimates Irreproducible Discovery Rates (IDR) for peaks in
#' replicated ChIP-seq experiments.
#'
#' @references
#' Q. Li, J. B. Brown, H. Huang and P. J. Bickel. (2011) Measuring
#' reproducibility of high-throughput experiments. Annals of Applied
#' Statistics, Vol. 5, No. 3, 1752-1779.
#'
#' @param remove_nonstandard_chromosomes removes peaks containing
#' genomic locations on non-standard chromosomes using
#' \code{\link[GenomeInfoDb:seqlevels-wrappers]{keepStandardChromosomes}}
#' (default is TRUE)
#' @param max_iteration integer; maximum number of iterations for
#' IDR estimation (defaults to 30)
#' @param local_idr see \code{\link[idr:est.IDR]{est.IDR}}
#'
#' @inheritParams establish_bijection1d
#' @inheritParams idr::est.IDR
#' @inheritParams preprocess
#'
#' @return List with three components, (\code{rep1_df}, \code{rep2_df},
#' and \code{analysis_type}) containing the interactions from input
#' data frames \code{rep1_df} and \code{rep2_df} with
#' the following additional columns:
#' \tabular{rll}{
#' column 1: \tab \code{chr} \tab character; genomic location of peak -
#' chromosome (e.g., \code{"chr3"})\cr
#' column 2: \tab \code{start} \tab integer; genomic location of peak -
#' start coordinate\cr
#' column 3: \tab \code{end} \tab integer; genomic location of peak -
#' end coordinate\cr
#' column 4: \tab \code{value} \tab numeric; p-value, FDR, or heuristic used
#' to rank the peaks\cr
#' column 5: \tab \code{rep_value} \tab numeric; value of corresponding
#' replicate peak. If no corresponding peak was found, \code{rep_value} is set
#' to \code{NA}.\cr
#' column 6: \tab \code{rank} \tab integer; rank of the peak, established by
#' value column, ascending order\cr
#' column 7: \tab \code{rep_rank} \tab integer; rank of corresponding
#' replicate peak. If no corresponding peak was found, \code{rep_rank} is
#' set to \code{NA}.\cr
#' column 8: \tab \code{idx} \tab integer; peak index, primary key\cr
#' column 9: \tab \code{rep_idx} \tab integer; specifies the index of the
#' corresponding peak in the other replicate (foreign key). If no
#' corresponding peak was found, \code{rep_idx} is set to \code{NA}.\cr
#' column 10: \tab \code{idr} \tab IDR of the peak and the
#' corresponding peak in the other replicate. If no corresponding
#' peak was found, \code{idr} is set to \code{NA}.
#' }
#'
#' @examples
#' idr_results <- estimate_idr1d(idr2d:::chipseq$rep1_df,
#' idr2d:::chipseq$rep2_df,
#' value_transformation = "log")
#' summary(idr_results)
#'
#' @export
estimate_idr1d <- function(rep1_df, rep2_df,
value_transformation = c("identity",
"additive_inverse",
"multiplicative_inverse",
"log",
"log_additive_inverse"),
ambiguity_resolution_method = c("overlap",
"midpoint",
"value"),
remove_nonstandard_chromosomes = TRUE,
max_factor = 1.5, jitter_factor = 0.0001,
max_gap = -1L,
mu = 0.1, sigma = 1.0, rho = 0.2, p = 0.5,
eps = 0.001, max_iteration = 30, local_idr = TRUE) {
return(estimate_idr(rep1_df, rep2_df, "IDR1D",
value_transformation,
ambiguity_resolution_method,
remove_nonstandard_chromosomes,
max_factor, jitter_factor, max_gap,
mu, sigma, rho, p,
eps, max_iteration, local_idr))
}
#' @title Estimates IDR for Genomic Interaction Data
#'
#' @description
#' This method estimates Irreproducible Discovery Rates (IDR) between
#' two replicates of experiments identifying genomic interactions, such as
#' Hi-C, ChIA-PET, and HiChIP.
#'
#' @references
#' Q. Li, J. B. Brown, H. Huang and P. J. Bickel. (2011) Measuring
#' reproducibility of high-throughput experiments. Annals of Applied
#' Statistics, Vol. 5, No. 3, 1752-1779.
#'
#' @param remove_nonstandard_chromosomes removes interactions
#' containing
#' genomic locations on non-standard chromosomes using
#' \code{\link[GenomeInfoDb:seqlevels-wrappers]{keepStandardChromosomes}}
#' (default is TRUE)
#' @param max_iteration integer; maximum number of iterations for
#' IDR estimation (defaults to 30)
#' @param local_idr see \code{\link[idr:est.IDR]{est.IDR}}
#'
#' @inheritParams establish_bijection2d
#' @inheritParams idr::est.IDR
#' @inheritParams preprocess
#'
#' @return List with three components, (\code{rep1_df}, \code{rep2_df},
#' and \code{analysis_type}) containing the interactions from input
#' data frames \code{rep1_df} and \code{rep2_df} with
#' the following additional columns:
#' \tabular{rll}{
#' column 1: \tab \code{chr_a} \tab character; genomic location of anchor A -
#' chromosome (e.g., \code{"chr3"})\cr
#' column 2: \tab \code{start_a} \tab integer; genomic location of anchor A -
#' start coordinate\cr
#' column 3: \tab \code{end_a} \tab integer; genomic location of anchor A -
#' end coordinate\cr
#' column 4: \tab \code{chr_b} \tab character; genomic location of anchor B -
#' chromosome (e.g., \code{"chr3"})\cr
#' column 5: \tab \code{start_b} \tab integer; genomic location of anchor B -
#' start coordinate\cr
#' column 6: \tab \code{end_b} \tab integer; genomic location of anchor B -
#' end coordinate\cr
#' column 7: \tab \code{value} \tab numeric; p-value, FDR, or heuristic used
#' to rank the interactions\cr
#' column 8: \tab \code{"rep_value"} \tab numeric; value of corresponding
#' replicate interaction. If no corresponding interaction was found,
#' \code{rep_value} is set to \code{NA}.\cr
#' column 9: \tab \code{"rank"} \tab integer; rank of the interaction,
#' established by value column, ascending order\cr
#' column 10: \tab \code{"rep_rank"} \tab integer; rank of corresponding
#' replicate interaction. If no corresponding interaction was found,
#' \code{rep_rank} is set to \code{NA}.\cr
#' column 11: \tab \code{"idx"} \tab integer; interaction index,
#' primary key\cr
#' column 12: \tab \code{"rep_idx"} \tab integer; specifies the index of the
#' corresponding interaction in the other replicate (foreign key). If no
#' corresponding interaction was found, \code{rep_idx} is set to \code{NA}.\cr
#' \code{idr} \tab IDR of the interaction and the
#' corresponding interaction in the other replicate. If no corresponding
#' interaction was found, \code{idr} is set to \code{NA}.
#' }
#'
#' @examples
#' idr_results <- estimate_idr2d(idr2d:::chiapet$rep1_df,
#' idr2d:::chiapet$rep2_df,
#' value_transformation = "log_additive_inverse")
#' summary(idr_results)
#'
#' @export
estimate_idr2d <- function(rep1_df, rep2_df,
value_transformation = c("identity",
"additive_inverse",
"multiplicative_inverse",
"log",
"log_additive_inverse"),
ambiguity_resolution_method = c("overlap",
"midpoint",
"value"),
remove_nonstandard_chromosomes = TRUE,
max_factor = 1.5, jitter_factor = 0.0001,
max_gap = -1L,
mu = 0.1, sigma = 1.0, rho = 0.2, p = 0.5,
eps = 0.001, max_iteration = 30, local_idr = TRUE) {
return(estimate_idr(rep1_df, rep2_df, "IDR2D",
value_transformation,
ambiguity_resolution_method,
remove_nonstandard_chromosomes,
max_factor, jitter_factor,
max_gap,
mu, sigma, rho, p,
eps, max_iteration, local_idr))
}
#' @title Estimates IDR for Genomic Peaks or Genomic Interactions
#' @references
#' Q. Li, J. B. Brown, H. Huang and P. J. Bickel. (2011) Measuring
#' reproducibility of high-throughput experiments. Annals of Applied
#' Statistics, Vol. 5, No. 3, 1752-1779.
#'
#' @param remove_nonstandard_chromosomes removes peaks and interactions
#' containing
#' genomic locations on non-standard chromosomes using
#' \code{\link[GenomeInfoDb:seqlevels-wrappers]{keepStandardChromosomes}}
#' (default is TRUE)
#' @param max_iteration integer; maximum number of iterations for
#' IDR estimation (defaults to 30)
#' @param local_idr see \code{\link[idr:est.IDR]{est.IDR}}
#'
#' @inheritParams idr::est.IDR
#' @inheritParams preprocess
#' @inheritParams establish_bijection
#'
#'
#' @return See \code{\link{estimate_idr1d}} or
#' \code{\link{estimate_idr2d}}, respectively.
#'
#' @examples
#' idr_results <- estimate_idr(idr2d:::chiapet$rep1_df,
#' idr2d:::chiapet$rep2_df,
#' analysis_type = "IDR2D",
#' value_transformation = "log_additive_inverse")
#' summary(idr_results)
#'
#' @importFrom dplyr arrange
#' @importFrom dplyr select
#' @importFrom dplyr filter
#' @importFrom futile.logger flog.warn
#' @importFrom stringr str_trim
#' @importFrom idr est.IDR
#' @export
estimate_idr <- function(rep1_df, rep2_df, analysis_type = "IDR2D",
value_transformation = c("identity",
"additive_inverse",
"multiplicative_inverse",
"log",
"log_additive_inverse"),
ambiguity_resolution_method = c("overlap",
"midpoint",
"value"),
remove_nonstandard_chromosomes = TRUE,
max_factor = 1.5, jitter_factor = 0.0001,
max_gap = -1L,
mu = 0.1, sigma = 1.0, rho = 0.2, p = 0.5,
eps = 0.001, max_iteration = 30, local_idr = TRUE) {
# avoid CRAN warnings
rep2_idx <- rep1_value <- rep2_value <- idr <- NULL
# argument handling
analysis_type <- match.arg(analysis_type, choices = c("IDR1D", "IDR2D"))
if (analysis_type == "IDR1D") {
columns <- c("chr", "start", "end", "value")
} else if (analysis_type == "IDR2D") {
columns <- c("chr_a", "start_a", "end_a",
"chr_b", "start_b", "end_b", "value")
}
if (nrow(rep1_df) > 0 && nrow(rep2_df) > 0) {
# remove irrelevant columns, rename relevant columns
rep1_df <- rep1_df[, seq_len(length(columns))]
rep2_df <- rep2_df[, seq_len(length(columns))]
colnames(rep1_df) <- columns
colnames(rep2_df) <- columns
if (remove_nonstandard_chromosomes) {
if (analysis_type == "IDR1D") {
rep1_df <- remove_nonstandard_chromosomes1d(rep1_df)
rep2_df <- remove_nonstandard_chromosomes1d(rep2_df)
} else if (analysis_type == "IDR2D") {
rep1_df <- remove_nonstandard_chromosomes2d(rep1_df)
rep2_df <- remove_nonstandard_chromosomes2d(rep2_df)
}
}
rep1_df$value <- preprocess(rep1_df$value,
value_transformation,
max_factor = max_factor,
jitter_factor = jitter_factor)
rep2_df$value <- preprocess(rep2_df$value,
value_transformation,
max_factor = max_factor,
jitter_factor = jitter_factor)
mapping <- establish_bijection(rep1_df, rep2_df, analysis_type,
ambiguity_resolution_method,
max_gap = max_gap)
if (nrow(mapping$rep1_df) > 0 && nrow(mapping$rep2_df) > 0) {
rep1_df <- mapping$rep1_df
rep2_df <- mapping$rep2_df
idx_df <- data.frame(
rep1_idx = rep1_df$idx,
rep2_idx = rep1_df$rep_idx,
rep1_value = rep1_df$value,
rep2_value = rep1_df$rep_value
)
idx_df <- dplyr::filter(idx_df,
!is.na(rep2_idx) & !is.infinite(rep2_idx))
if (nrow(idx_df) > 0) {
idr_matrix <- as.matrix(dplyr::select(idx_df,
rep1_value,
rep2_value))
if (length(unique(idr_matrix[, 1])) < 10 ||
length(unique(idr_matrix[, 2])) < 10) {
futile.logger::flog.warn("low complexity data set")
}
invisible(tryCatch({
idr_results <- idr::est.IDR(idr_matrix, mu, sigma, rho, p,
eps = eps,
max.ite = max_iteration)
if (local_idr) {
idx_df$idr <- idr_results$idr
} else {
idx_df$idr <- idr_results$IDR
}
},
error = function(e) {
idx_df$idr <- as.numeric(NA)
futile.logger::flog.warn(stringr::str_trim(e))
}))
} else {
idx_df$idr <- numeric(0)
}
if (nrow(rep1_df) > 0) {
rep1_df$idr <- as.numeric(NA)
if (nrow(idx_df) > 0) {
rep1_df$idr[idx_df$rep1_idx] <- idx_df$idr
rep1_df <- dplyr::arrange(rep1_df, idr)
}
} else {
rep1_df$idr <- numeric(0)
}
if (nrow(rep2_df) > 0) {
rep2_df$idr <- as.numeric(NA)
if (length(idx_df$idr) > 0) {
rep2_df$idr[idx_df$rep2_idx] <- idx_df$idr
rep2_df <- dplyr::arrange(rep2_df, idr)
}
} else {
rep2_df$idr <- numeric(0)
}
} else {
if (nrow(rep1_df) > 0) {
rep1_df$idr <- as.numeric(NA)
} else {
rep1_df$idr <- numeric(0)
}
if (nrow(rep2_df) > 0) {
rep2_df$idr <- as.numeric(NA)
} else {
rep2_df$idr <- numeric(0)
}
}
columns <- c(columns, "rep_value", "rank", "rep_rank",
"idx", "rep_idx", "idr")
if (analysis_type == "IDR1D") {
rep1_df <- dplyr::select(rep1_df, columns)
rep2_df <- dplyr::select(rep2_df, columns)
} else if (analysis_type == "IDR2D") {
rep1_df <- dplyr::select(rep1_df, columns)
rep2_df <- dplyr::select(rep2_df, columns)
}
} else {
if (nrow(rep1_df) == 0) {
rep1_df$idx <- integer(0)
rep1_df$rep_idx <- integer(0)
rep1_df$rank <- integer(0)
rep1_df$rep_rank <- integer(0)
rep1_df$rep_value <- numeric(0)
rep1_df$idr <- numeric(0)
} else {
# remove irrelevant columns, rename relevant columns
rep1_df <- rep1_df[, seq_len(length(columns))]
colnames(rep1_df) <- columns
columns <- c(columns, "rep_value", "rank",
"rep_rank", "idx", "rep_idx", "idr")
rep1_df$idx <- NA_integer_
rep1_df$rep_idx <- NA_integer_
rep1_df$rank <- NA_integer_
rep1_df$rep_rank <- NA_integer_
rep1_df$rep_value <- NA_real_
rep1_df$idr <- NA_real_
rep1_df <- dplyr::select(rep1_df, columns)
}
if (nrow(rep2_df) == 0) {
rep2_df$idx <- integer(0)
rep2_df$rep_idx <- integer(0)
rep2_df$rank <- integer(0)
rep2_df$rep_rank <- integer(0)
rep2_df$rep_value <- numeric(0)
rep2_df$idr <- numeric(0)
} else {
# remove irrelevant columns, rename relevant columns
rep2_df <- rep2_df[, seq_len(length(columns))]
colnames(rep2_df) <- columns
columns <- c(columns, "rep_value", "rank",
"rep_rank", "idx", "rep_idx", "idr")
rep2_df$idx <- NA_integer_
rep2_df$rep_idx <- NA_integer_
rep2_df$rank <- NA_integer_
rep2_df$rep_rank <- NA_integer_
rep2_df$rep_value <- NA_real_
rep2_df$idr <- NA_real_
rep2_df <- dplyr::select(rep2_df, columns)
}
}
res <- list(rep1_df = rep1_df, rep2_df = rep2_df,
analysis_type = analysis_type)
class(res) <- c("idr2d_result", class(res))
return(res)
}
#' @importFrom methods is
#' @importFrom dplyr filter
#' @export
summary.idr2d_result <- function(object, ...) {
idr <- NULL
stopifnot(methods::is(object, "idr2d_result"))
num_rep_df <- dplyr::filter(object$rep1_df, !is.na(idr))
num_sig_df <- dplyr::filter(num_rep_df, idr < 0.05)
num_high_sig_df <- dplyr::filter(num_rep_df, idr < 0.01)
res <- list(
analysis_type = object$analysis_type,
rep1_num_interactions = nrow(object$rep1_df),
rep2_num_interactions = nrow(object$rep2_df),
num_reproducible_interactions = nrow(num_rep_df),
num_significant_interactions = nrow(num_sig_df),
num_highly_significant_interactions = nrow(num_high_sig_df)
)
class(res) <- c("idr2d_result_summary", class(res))
return(res)
}
#' @importFrom methods is
#' @export
print.idr2d_result_summary <- function(x, ...) {
stopifnot(methods::is(x, "idr2d_result_summary"))
cat("analysis type: ", x$analysis_type, "\n", sep = "")
if (x$analysis_type == "IDR2D Hi-C") {
cat("number of blocks: ",
x$num_blocks, "\n", sep = "")
cat("number of blocks with significant IDR (IDR < 0.05): ",
x$num_significant_blocks, "\n", sep = "")
cat("number of blocks with highly significant IDR (IDR < 0.01): ",
x$num_highly_significant_blocks, "\n", sep = "")
cat("percentage of blocks with significant IDR (IDR < 0.05): ",
round(x$num_significant_blocks / x$num_blocks * 100, digits = 2),
" %\n", sep = "")
} else {
cat("number of interactions in replicate 1: ",
x$rep1_num_interactions, "\n", sep = "")
cat("number of interactions in replicate 2: ",
x$rep2_num_interactions, "\n", sep = "")
cat("number of reproducible interactions: ",
x$num_reproducible_interactions, "\n", sep = "")
cat("number of interactions with significant IDR (IDR < 0.05): ",
x$num_significant_interactions, "\n", sep = "")
cat("number of interactions with highly significant IDR (IDR < 0.01): ",
x$num_highly_significant_interactions, "\n", sep = "")
cat("percentage of interactions with significant IDR (IDR < 0.05): ",
round(x$num_significant_interactions /
max(x$rep1_num_interactions, x$rep2_num_interactions) * 100,
digits = 2),
" %\n", sep = "")
}
}
Try the idr2d package in your browser
Any scripts or data that you put into this service are public.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.