Mercurial > repos > recetox > recetox_aplcms_align_features
view utils.R @ 0:57c644d3f24c draft
planemo upload for repository https://github.com/RECETOX/galaxytools/tree/master/tools/recetox_aplcms commit 19de0924a65bc65cbbf7c1fc17e9b5348305f95c
author | recetox |
---|---|
date | Fri, 10 Jun 2022 10:17:46 +0000 |
parents | |
children | abe783e0daca |
line wrap: on
line source
library(recetox.aplcms) align_features <- function(sample_names, ...) { aligned <- feature.align(...) feature_names <- seq_len(nrow(aligned$pk.times)) list( mz_tolerance = as.numeric(aligned$mz.tol), rt_tolerance = as.numeric(aligned$chr.tol), rt_crosstab = as_feature_crosstab(feature_names, sample_names, aligned$pk.times), int_crosstab = as_feature_crosstab(feature_names, sample_names, aligned$aligned.ftrs) ) } get_sample_name <- function(filename) { tools::file_path_sans_ext(basename(filename)) } as_feature_crosstab <- function(feature_names, sample_names, data) { colnames(data) <- c("mz", "rt", "mz_min", "mz_max", sample_names) rownames(data) <- feature_names as.data.frame(data) } as_feature_sample_table <- function(rt_crosstab, int_crosstab) { feature_names <- rownames(rt_crosstab) sample_names <- colnames(rt_crosstab)[- (1:4)] feature_table <- data.frame( feature = feature_names, mz = rt_crosstab[, 1], rt = rt_crosstab[, 2] ) # series of conversions to produce a table type from data.frame rt_crosstab <- as.table(as.matrix(rt_crosstab[, - (1:4)])) int_crosstab <- as.table(as.matrix(int_crosstab[, - (1:4)])) crosstab_axes <- list(feature = feature_names, sample = sample_names) dimnames(rt_crosstab) <- dimnames(int_crosstab) <- crosstab_axes x <- as.data.frame(rt_crosstab, responseName = "sample_rt") y <- as.data.frame(int_crosstab, responseName = "sample_intensity") data <- merge(x, y, by = c("feature", "sample")) data <- merge(feature_table, data, by = "feature") data } load_features <- function(files) { files_list <- sort_samples_by_acquisition_number(files) features <- lapply(files_list, arrow::read_parquet) features <- lapply(features, as.matrix) return(features) } save_data_as_parquet_files <- function(data, subdir) { dir.create(subdir) for (i in 0:(length(data) - 1)) { filename <- file.path(subdir, paste0(subdir, "_features_", i, ".parquet")) arrow::write_parquet(as.data.frame(data[i + 1]), filename) } } save_aligned_features <- function(aligned, rt_file, int_file, tol_file) { arrow::write_parquet(as.data.frame(aligned$rt_crosstab), rt_file) arrow::write_parquet(as.data.frame(aligned$int_crosstab), int_file) mz_tolerance <- c(aligned$mz_tolerance) rt_tolerance <- c(aligned$rt_tolerance) arrow::write_parquet(data.frame(mz_tolerance, rt_tolerance), tol_file) } load_aligned_features <- function(rt_file, int_file, tol_file) { rt_cross_table <- arrow::read_parquet(rt_file) int_cross_table <- arrow::read_parquet(int_file) tolerances_table <- arrow::read_parquet(tol_file) result <- list() result$mz_tolerance <- tolerances_table$mz_tolerance result$rt_tolerance <- tolerances_table$rt_tolerance result$rt_crosstab <- rt_cross_table result$int_crosstab <- int_cross_table return(result) } recover_signals <- function(cluster, filenames, extracted, corrected, aligned, mz_tol = 1e-05, mz_range = NA, rt_range = NA, use_observed_range = TRUE, min_bandwidth = NA, max_bandwidth = NA, recover_min_count = 3) { if (!is(cluster, "cluster")) { cluster <- parallel::makeCluster(cluster) on.exit(parallel::stopCluster(cluster)) } clusterExport(cluster, c("extracted", "corrected", "aligned", "recover.weaker")) clusterEvalQ(cluster, library("splines")) recovered <- parLapply(cluster, seq_along(filenames), function(i) { recover.weaker( loc = i, filename = filenames[[i]], this.f1 = extracted[[i]], this.f2 = corrected[[i]], pk.times = aligned$rt_crosstab, aligned.ftrs = aligned$int_crosstab, orig.tol = mz_tol, align.mz.tol = aligned$mz_tolerance, align.chr.tol = aligned$rt_tolerance, mz.range = mz_range, chr.range = rt_range, use.observed.range = use_observed_range, bandwidth = 0.5, min.bw = min_bandwidth, max.bw = max_bandwidth, recover.min.count = recover_min_count ) }) feature_table <- aligned$rt_crosstab[, 1:4] rt_crosstab <- cbind(feature_table, sapply(recovered, function(x) x$this.times)) int_crosstab <- cbind(feature_table, sapply(recovered, function(x) x$this.ftrs)) feature_names <- rownames(feature_table) sample_names <- colnames(aligned$rt_crosstab[, - (1:4)]) list( extracted_features = lapply(recovered, function(x) x$this.f1), corrected_features = lapply(recovered, function(x) x$this.f2), rt_crosstab = as_feature_crosstab(feature_names, sample_names, rt_crosstab), int_crosstab = as_feature_crosstab(feature_names, sample_names, int_crosstab) ) } create_feature_sample_table <- function(features) { table <- as_feature_sample_table( rt_crosstab = features$rt_crosstab, int_crosstab = features$int_crosstab ) return(table) }