############################################################################## #============================================================================ # Framework to find microbial signatures of Pulmonary Exacerbation in CF #============================================================================ # # R codes to find shifts in relative abundance of keystone taxa that # correspond to changes in patient health. # # Manuscript: Microbiome Networks and Change-Point Analysis Reveal Key # Community Changes Associated with Cystic Fibrosis Pulmonary Exacerbations # # Journal: npj Biofilms and Microbiomes # # Author: Mehdi Layeghifard et al. # # Date: November/2018 # ##### Install Required packages # install.packages("igraph") # install.packages("qgraph") # install.packages("changepoint") # # Also requires function in the utils.R file # ##### Input data # A comma-separated OTU table (.csv file) for each patient with rows and # columns representing taxa and samples, respectively. # # A comma-separated metadata file (.csv) containing data on clinical state # of each sample for each patient. Rows and columns represent samples and # patients, respectively. # ##### Disclaimer # This code is supplied without any warranty or guaranteed support whatsoever. # Authors are not responsible for its use, misuse, or functionality. # #============================================================================ # Contact: Mehdi Layeghifard (mehdi.layeghifard@utoronto.ca) #============================================================================ ############################################################################## # Load libraries library(igraph) library(qgraph) library(changepoint) # Load custom functions needed to run the framework source("utils.R") options(warn=-1) ##### Define parameters # Path to data folder data_path <- "data" # Number of desired keystone taxa top <- 5 # The hub detection method: pagerank (default) or hubscore hub_metric <- "pagerank" # FDR network metric: see qgraph manual for more detail fdrm <- "pval" # Define containers for keystone taxa chpt_bugs <- list() hubs_bugs <- list() abun_bugs <- list() prev_bugs <- list() # Define containers for covariance measures chpt_covs <- list() hubs_covs <- list() abun_covs <- list() prev_covs <- list() # Get the list of otu files files <- list.files(path = data_path, pattern = "_otu.csv") # Order file names files <- files[order(nchar(files), files)] # Read metadata (clinical states) metadata <- read.table(paste(data_path, "clinical_states.csv", sep="/"), header=TRUE, sep=",") # Compute changepoints and covariances for each otu table for (fname in files) { patient_id <- strsplit(fname, "_")[[1]][1] writeLines(paste("Processing: ", patient_id)) # Read the otu file infile <- paste(data_path, fname, sep="/") MB <- read.table(infile, header=TRUE, sep=",") num_taxa <- dim(MB)[1] num_samples <- dim(MB)[2] # Create the FDR network using qgraph package CorMat <- cor_auto(t(MB), verbose = FALSE) PCorMat_FDR <- FDRnetwork(CorMat, cutoff = 0.01, method = fdrm) FDRnet <- qgraph(PCorMat_FDR, DoNotPlot = TRUE) net <- q2igraph(FDRnet) # Find changepoints in clinical states clin_state <- metadata[[patient_id]] clin_state <- clin_state[!is.na(clin_state)] cp1 <- cpt.meanvar(clin_state, method="PELT") cs_vec1 <- as.vector(matrix(0, nrow=dim(MB)[2])) ####### j <- 1 for (i in 1:length(cp1@cpts)) { cs_vec1[j:cp1@cpts[i]] <- i j <- cp1@cpts[i] + 1 } # Find keystone taxa abun <- names(head(sort(rowSums(MB), decreasing=TRUE), top)) prev <- names(head(sort(rowSums(MB != 0), decreasing=TRUE), top)) hubs <- find_hubs(net, hub_metric="pagerank") # Extract relative abundance of keystone taxa from otu table abun_mat <- MB[abun, ] prev_mat <- MB[prev, ] hubs_mat <- MB[hubs, ] # Define containers to store covariance measures for each otu chpt_res <- array(0, dim=c(3, num_taxa)) colnames(chpt_res) <- rownames(MB) abun_res <- array(0, dim=c(3, top)) prev_res <- array(0, dim=c(3, top)) hubs_res <- array(0, dim=c(3, top)) # Find changepoint covariances for all taxa for (i in 1:num_taxa) { chpt_cpt <- cpt.meanvar(as.numeric(MB[i,]), method="PELT") chpt_vec <- cpt_conv(chpt_cpt, num_samples) chpt_ccf <- ccf(chpt_vec, cs_vec1, lag.max = 1, type="covariance", plot=F) chpt_res[, i] <- chpt_ccf$acf } # Select the top taxa chpt_res <- chpt_res[, order(chpt_res[2,], decreasing=T)] chpt_res <- chpt_res[, 1:top] # Find changepoint covariance for keystone taxa for (i in 1:top) { abun_cpt <- cpt.meanvar(as.numeric(abun_mat[i,]), method="PELT") prev_cpt <- cpt.meanvar(as.numeric(prev_mat[i,]), method="PELT") hubs_cpt <- cpt.meanvar(as.numeric(hubs_mat[i,]), method="PELT") abun_vec <- cpt_conv(abun_cpt, num_samples) prev_vec <- cpt_conv(prev_cpt, num_samples) hubs_vec <- cpt_conv(hubs_cpt, num_samples) abun_ccf <- ccf(abun_vec, cs_vec1, lag.max = 1, type="covariance", plot=F) prev_ccf <- ccf(prev_vec, cs_vec1, lag.max = 1, type="covariance", plot=F) hubs_ccf <- ccf(hubs_vec, cs_vec1, lag.max = 1, type="covariance", plot=F) abun_res[, i] <- abun_ccf$acf prev_res[, i] <- prev_ccf$acf hubs_res[, i] <- hubs_ccf$acf } # Store keystone taxa found for this otu chpt_bugs[[patient_id]] <- colnames(chpt_res) hubs_bugs[[patient_id]] <- hubs abun_bugs[[patient_id]] <- abun prev_bugs[[patient_id]] <- prev # Store covariances of keystone taxa found for this otu chpt_covs[[patient_id]] <- cumsum(abs(chpt_res[2,])) hubs_covs[[patient_id]] <- cumsum(abs(hubs_res[2,])) abun_covs[[patient_id]] <- cumsum(abs(abun_res[2,])) prev_covs[[patient_id]] <- cumsum(abs(prev_res[2,])) } # Convert lists of keystone taxa to dataframes chpt_df <- data.frame(chpt_bugs) abun_df <- data.frame(abun_bugs) prev_df <- data.frame(prev_bugs) hubs_df <- data.frame(hubs_bugs) # Add row names rownames(chpt_df) <- paste0("top.", c(1:top)) rownames(abun_df) <- paste0("top.", c(1:top)) rownames(prev_df) <- paste0("top.", c(1:top)) rownames(hubs_df) <- paste0("top.", c(1:top)) # Convert lists of covariance measures to matrices chpt_covs <- as.matrix(data.frame(chpt_covs)) hubs_covs <- as.matrix(data.frame(hubs_covs)) abun_covs <- as.matrix(data.frame(abun_covs)) prev_covs <- as.matrix(data.frame(prev_covs)) # Combine all covariance measures into one data frame all_covs <- cbind(abun_covs, prev_covs, hubs_covs) # Siginificance test significance_res <- wilcoxtest(all_covs, top) writeLines("\n\nResults of the significance test:\n") print(significance_res) #### Write the results to files # Write keystone taxa write.table(chpt_df, "keystone_changepoint.csv", sep=",") write.table(abun_df, "keystone_abundance.csv", sep=",") write.table(prev_df, "keystone_prevalence.csv", sep=",") write.table(hubs_df, "keystone_hubs.csv", sep=",") # Write covariance measures write.table(chpt_covs, "covariances_changepoint.csv", sep=",") write.table(abun_covs, "covariances_abundance.csv", sep=",") write.table(prev_covs, "covariances_prevalence.csv", sep=",") write.table(hubs_covs, "covariances_hubs.csv", sep=",") # Write significance test results write.table(significance_res, "significance_results.csv", sep=",")