Mercurial > repos > bornea > saint_bubblebeam
diff bubbles_v9_NSAF_natural_log.R @ 25:ab602bbf4ac5 draft
Uploaded
author | bornea |
---|---|
date | Fri, 29 Jan 2016 09:33:58 -0500 |
parents | |
children |
line wrap: on
line diff
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/bubbles_v9_NSAF_natural_log.R Fri Jan 29 09:33:58 2016 -0500 @@ -0,0 +1,268 @@ +rm(list = ls()) +################################################################################################### +# R-code: Multi-bubble graph generation from SAINTexpress output +# Author: Brent Kuenzi +################################################################################################### +# This Script generates the bubble graphs based upon Saint output. +################################################################################################### +# Copyright (C) Brent Kuenzi. +# Permission is granted to copy, distribute and/or modify this document +# under the terms of the GNU Free Documentation License, Version 1.3 +# or any later version published by the Free Software Foundation; +# with no Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts. +# A copy of the license is included in the section entitled "GNU +# Free Documentation License". +################################################################################################### +## REQUIRED INPUT ## + +# 1) listfile: SAINTexpress generated "list.txt" file +# 2) preyfile: SAINT pre-processing generated "prey.txt" file used to run SAINTexpress + +## OPTIONAL INPUT ## + +# 3) crapome: raw output from crapome Workflow 1 query (http://www.crapome.org) +# 4) color: bubble color (default = "red") +# - color = "crapome": color bubbles based on Crapome(%) +# - Also recognizes any color within R's built-in colors() vector +# 5) label: Adds gene name labels to bubbles within the "zoomed in" graphs (default = FALSE) +# 6) cutoff: Saintscore cutoff to be assigned for filtering the "zoomed in" graphs (default = 0.8) +# 7) type: Specifies if the data is MaxQuant (MQ) or Scaffold (SC) data (default = "SC") +# 8) inc_file: Selects only the uniprot ids in the provided list (default ="None") +# 9) exc_file: Removes the proteins in the list (default = "None") +################################################################################################### + + +ins_check_run <- function(){ + if ('dplyr' %in% rownames(installed.packages())){} + else { + install.packages('dplyr', repos = 'http://cran.us.r-project.org') + } + if ('tidyr' %in% rownames(installed.packages())){} + else { + install.packages('tidyr', repos = 'http://cran.us.r-project.org') + } + if ('ggplot2' %in% rownames(installed.packages())){} + else { + install.packages('ggplot2', repos = 'http://cran.us.r-project.org') + } +} + +ins_check_run() +library(dplyr); library(tidyr); library(ggplot2) + + +main <- function(listfile, preyfile, crapome = FALSE, color = "red", label = FALSE, cutoff = 0.8, type = "SC", inc_file = "None", exc_file = "None" ) { + cutoff_check(cutoff) + listfile <- list_type(listfile, inc_file, exc_file) + if(type == "SC") { + df <- merge_files_sc(listfile, preyfile, crapome) + } + if(type == "MQ") { + df <- merge_files_mq(listfile, preyfile, crapome) + } + bubble_NSAF(df, color) + bubble_SAINT(df, color) + bubble_zoom_SAINT(df, color, label, cutoff) + bubble_zoom_NSAF(df, color, label, cutoff) + write.table(df, "output.txt", sep = "\t", quote = FALSE, row.names = FALSE) +} +################################################################################################### +# Include and Exclude list filtering +################################################################################################### +list_type <- function(df, inc_file, exc_file) { + Saint <- read.delim(df, stringsAsFactors = FALSE) + if (inc_file != "None") { + if (exc_file == "None") { + inc_prots <- read.delim(inc_file, sep = '\t', header = FALSE, stringsAsFactors = FALSE) + print(inc_prots[, 1]) + print(Saint$Prey) + filtered_df = subset(Saint, Saint$Prey == inc_prots[, 1]) + } + else { + inc_prots <- read.delim(inc_file, sep = '\t', header = FALSE, stringsAsFactors = FALSE) + exc_prots <- read.delim(exc_file, sep = '\t', header = FALSE, stringsAsFactors = FALSE) + filtered_df = subset(Saint, Saint$Prey == inc_prots[, 1]) + filtered_df = subset(filtered_df, filtered_df$Prey != exc_prots[, 1]) + } + } + else if (exc_file != "None") { + exc_prots <- read.delim(exc_file, sep = '\t', header = FALSE, stringsAsFactors = FALSE) + filtered_df = subset(Saint, Saint$Prey != exc_prots[, 1]) + } + else { + filtered_df = Saint + } + return(filtered_df) + +} +################################################################################################### +# Merge input files and caculate Crapome(%) and NSAF for each protein for each bait +################################################################################################### +merge_files_mq <- function(SAINT, prey_DF, crapome = FALSE) { + #SAINT <- read.table(SAINT_DF, sep = '\t', header = TRUE) + #Some of these read.table()'s don't use stringsAsFactors = FALSE. Is this on purpose? Factors give rise to some really weird and unpredictable behavior; suggest always using stringsAsFactors = FALSE + prey <- read.table(prey_DF, sep = '\t', header = FALSE); colnames(prey) <- c("Prey", "Length", "PreyGene") + DF <- merge(SAINT, prey) + DF$SpecSum <- log2(DF$SpecSum) + + if(crapome != FALSE) { + crapome <- read.table(crapome, sep = '\t', header = TRUE) + colnames(crapome) <- c("Prey", "Symbol", "Num.of.Exp", "Ave.SC", "Max.SC") + DF1 <- merge(DF, crapome); as.character(DF1$Num.of.Exp); DF1$Symbol <- NULL; + DF1$Ave.SC <- NULL; DF1$Max.SC <- NULL # Removes unnecessary columns. + DF1$Num.of.Exp <- sub("^$", "0 / 1", DF1$Num.of.Exp ) # Replace blank values with 0 / 1. + DF <- DF1 %>% separate(Num.of.Exp, c("NumExp", "TotalExp"), " / ") # Split into 2 columns. + DF$CrapomePCT <- 100 - (as.integer(DF$NumExp) / as.integer(DF$TotalExp) * 100) # Calculate the crapome %. + } + DF$SAF <- DF$AvgSpec / DF$Length + DF2 = DF %>% group_by(Bait) %>% mutate(NSAF = SAF/sum(SAF)) + DF$NSAF = DF2$NSAF + return(DF) +} + +merge_files_sc <- function(SAINT, prey_DF, crapome = FALSE) { + #SAINT <- read.table(SAINT_DF, sep = '\t', header = TRUE) + prey <- read.table(prey_DF, sep = '\t', header = FALSE); colnames(prey) <- c("Prey", "Length", "PreyGene") + DF <- merge(SAINT, prey) + + if(crapome != FALSE) { + crapome <- read.table(crapome, sep = '\t', header = TRUE) + colnames(crapome) <- c("Prey", "Symbol", "Num.of.Exp", "Ave.SC", "Max.SC") + DF1 <- merge(DF, crapome); as.character(DF1$Num.of.Exp); DF1$Symbol <- NULL; + DF1$Ave.SC <- NULL; DF1$Max.SC <- NULL # Removes unnecessary columns. + DF1$Num.of.Exp <- sub("^$", "0 / 1", DF1$Num.of.Exp ) # Replace blank values with 0 / 1. + DF <- DF1 %>% separate(Num.of.Exp, c("NumExp", "TotalExp"), " / ") # Split into 2 columns. + DF$CrapomePCT <- 100 - (as.integer(DF$NumExp) / as.integer(DF$TotalExp) * 100) # Calculate the crapome %. + } + DF$SAF <- DF$AvgSpec / DF$Length + DF2 = DF %>% group_by(Bait) %>% mutate(NSAF = SAF/sum(SAF)) + DF$NSAF = DF2$NSAF + return(DF) +} +################################################################################################### +# Plot all proteins for each bait by x = ln(NSAF), y = Log2(FoldChange) +################################################################################################### +bubble_NSAF <- function(data, color) { + if(color == "crapome") { + a <- subset(data, CrapomePCT < 80, select = c(NSAF, SpecSum, CrapomePCT, FoldChange, SaintScore, Bait)) + b <- subset(data, CrapomePCT >= 80, select = c(NSAF, SpecSum, CrapomePCT, FoldChange, SaintScore, Bait)) + p <- qplot(x = log(NSAF), y = log2(FoldChange), data = a, colour = I("tan"), size = SpecSum) + scale_size(range = c(1, 10)) + + geom_point(aes(x = log(NSAF), y = log2(FoldChange), size = SpecSum), colour = "black", shape = 21, data = a) + if(length(levels(a$Bait) > 1)) {p <- p + facet_wrap(~Bait, scales = "free_y")} # multiple graphs if multiple baits + #The text says ln() which is log base e, but the code uses log base 10. Fix code or the axis label. + p <- p + geom_point(aes(x = log(NSAF), y = log2(FoldChange), size = SpecSum, color = CrapomePCT), data = b) + + scale_colour_gradient(limits = c(80, 100), low = "tan", high = "red") + + labs(colour = "CRAPome Probability \nof Specific Interaction (%)", x = "ln(NSAF)") + + geom_point(aes(x = log(NSAF), y = log2(FoldChange), size = SpecSum), colour = "black", shape = 21, data = b) + return(ggsave(p, width = 8, height = 4, filename = "bubble_NSAF.png")) + } + if(color != "crapome") { + p <- qplot(x = log(NSAF), y = log2(FoldChange), data = data, colour = I(color), size = SpecSum) + scale_size(range = c(1, 10)) + + geom_point(aes(x = log(NSAF), y = log2(FoldChange), size = SpecSum), colour = "black", shape = 21, data = data) + # add bubble outlines + labs(x = "ln(NSAF)") + if(length(levels(data$Bait) > 1)) {p <- p + facet_wrap(~Bait, scales = "free_y")} + return(ggsave(p, width = 8, height = 4, filename = "bubble_NSAF.png")) + } + } +################################################################################################### +# Plot all proteins for each bait by x = Saintscore, y = Log2(FoldChange) +################################################################################################### +bubble_SAINT <- function(data, color) { + if(color == "crapome") { + a <- subset(data, CrapomePCT < 80, select = c(NSAF, SpecSum, CrapomePCT, FoldChange, SaintScore, Bait)) #filter on CRAPome + b <- subset(data, CrapomePCT >= 80, select = c(NSAF, SpecSum, CrapomePCT, FoldChange, SaintScore, Bait)) + p <- qplot(x = SaintScore, y = log2(FoldChange), data = a, colour = I("tan"), size = SpecSum) + + scale_size(range = c(1, 10)) + geom_point(aes(x = SaintScore, y = log2(FoldChange), size = SpecSum), colour = "black", shape = 21, data = a) + if(length(levels(a$Bait) > 1)) {p <- p + facet_wrap(~Bait, scales = "free_y")} + p <- p + geom_point(aes(x = SaintScore, y = log2(FoldChange), size = SpecSum, color = CrapomePCT), data = b) + + scale_colour_gradient(limits = c(80, 100), low = "tan", high = "red") + + labs(colour = "CRAPome Probability \nof Specific Interaction (%)") + + geom_point(aes(x = SaintScore, y = log2(FoldChange), size = SpecSum), colour = "black", shape = 21, data = b) + return(ggsave(p, width = 8, height = 4, filename = "bubble_SAINT.png")) + } + if(color != "crapome") { + p <- qplot(x = SaintScore, y = log2(FoldChange), data = data, colour = I(color), size = SpecSum) + + scale_size(range = c(1, 10)) + geom_point(aes(x = SaintScore, y = log2(FoldChange), size = SpecSum), colour = "black", shape = 21, data = data) + if(length(levels(data$Bait) > 1)) {p <- p + facet_wrap(~Bait, scales = "free_y")} + return(ggsave(p, width = 8, height = 4, filename = "bubble_SAINT.png")) + } + } +################################################################################################### +# Filter proteins on Saintscore cutoff and plot for each bait x = Saintscore, y = Log2(FoldChange) +################################################################################################### +bubble_zoom_SAINT <- function(data, color, label = FALSE, cutoff = 0.8) { + if(color == "crapome") { + a <- subset(data, CrapomePCT < 80 & SaintScore >= cutoff, select = c(NSAF, SpecSum, CrapomePCT, FoldChange, SaintScore, Bait, PreyGene)) + b <- subset(data, CrapomePCT >= 80 & SaintScore >= cutoff, select = c(NSAF, SpecSum, CrapomePCT, FoldChange, SaintScore, Bait, PreyGene)) + p <- qplot(x = SaintScore, y = log2(FoldChange), data = a, colour = I("tan"), size = SpecSum) + + scale_size(range = c(1, 10)) + ggtitle("Filtered on SAINT score")+geom_point(aes(x = SaintScore, y = log2(FoldChange), size = SpecSum), colour = "black", shape = 21, data = a) + if(label == TRUE & length(a$NSAF != 0)) { + p <- p + geom_text(data = a, aes(label = PreyGene, size = 10, vjust = 0, hjust = 0), colour = "black") + } + if(length(levels(a$Bait) > 1)) {p <- p + facet_wrap(~Bait, scales = "free_y")} + p <- p + geom_point(aes(x = SaintScore, y = log2(FoldChange), size = SpecSum, color = CrapomePCT), data = b) + + scale_colour_gradient(limits = c(80, 100), low = "tan", high = "red") + + labs(colour = "CRAPome Probability \nof Specific Interaction (%)") + + geom_point(aes(x = SaintScore, y = log2(FoldChange), size = SpecSum), colour = "black", shape = 21, data = b) + if(label == TRUE & length(b$NSAF != 0)) { + p <- p + geom_text(data = b, aes(label = PreyGene, size = 10, vjust = 0, hjust = 0), colour = "black", show_guide = FALSE) + } + return(ggsave(p, width = 8, height = 4, filename = "bubble_zoom_SAINT.png")) + } + if(color != "crapome") { + a <- subset(data, SaintScore >= cutoff, select = c(NSAF, SpecSum, FoldChange, SaintScore, Bait, PreyGene)) + p <- qplot(x = SaintScore, y = log2(FoldChange), data = a, colour = I(color), size = SpecSum) + + scale_size(range = c(1, 10)) + ggtitle("Filtered on SAINT score") + + geom_point(aes(x = SaintScore, y = log2(FoldChange), size = SpecSum), colour = "black", shape = 21, data = a) + if(label == TRUE & length(a$NSAF != 0)) { + p <- p + geom_text(data = a, aes(label = PreyGene, size = 10, vjust = 0, hjust = 0), colour = "black", show_guide = FALSE) + } + if(length(levels(data$Bait) > 1)) {p <- p + facet_wrap(~Bait, scales = "free_y")} + return(ggsave(p, width = 8, height = 4, filename = "bubble_zoom_SAINT.png")) + } +} +################################################################################################### +# Filter proteins on Saintscore cutoff and plot for each bait x = log(NSAF), y = Log2(FoldChange) +################################################################################################### +bubble_zoom_NSAF <- function(data, color, label = FALSE, cutoff = 0.8) { + if(color == "crapome") { + a <- subset(data, CrapomePCT < 80 & SaintScore >= cutoff, select = c(NSAF, SpecSum, CrapomePCT, FoldChange, SaintScore, Bait, PreyGene)) + b <- subset(data, CrapomePCT >= 80 & SaintScore >= cutoff, select = c(NSAF, SpecSum, CrapomePCT, FoldChange, SaintScore, Bait, PreyGene)) + p <- qplot(x = log(NSAF), y = log2(FoldChange), data = a, colour = I("tan"), size = SpecSum) + + scale_size(range = c(1, 10)) + ggtitle("Filtered on SAINT score") + + geom_point(aes(x = log(NSAF), y = log2(FoldChange), size = SpecSum), colour = "black", shape = 21, data = a) + if(label == TRUE & length(a$NSAF != 0)) { + p <- p + geom_text(data = a, aes(label = PreyGene, size = 10, vjust = 0, hjust = 0), colour = "black") + } + if(length(levels(a$Bait) > 1)) {p <- p + facet_wrap(~Bait, scales = "free_y")} + p <- p + geom_point(aes(x = log(NSAF), y = log2(FoldChange), size = SpecSum, color = CrapomePCT), data = b) + + scale_colour_gradient(limits = c(80, 100), low = "tan", high = "red") + + labs(colour = "CRAPome Probability \nof Specific Interaction (%)", x = "ln(NSAF)") + + geom_point(aes(x = log(NSAF), y = log2(FoldChange), size = SpecSum), colour = "black", shape = 21, data = b) + if(label == TRUE & length(b$NSAF != 0)) { + p <- p + geom_text(data = b, aes(label = PreyGene, size = 10, vjust = 0, hjust = 0), colour = "black", show_guide = FALSE) + } + return(ggsave(p, width = 8, height = 4, filename = "bubble_zoom_NSAF.png")) + } + if(color != "crapome") { + a <- subset(data, SaintScore >= cutoff, select = c(NSAF, SpecSum, FoldChange, SaintScore, Bait, PreyGene)) + p <- qplot(x = log(NSAF), y = log2(FoldChange), data = a, colour = I(color), size = SpecSum) + + scale_size(range = c(1, 10)) + ggtitle("Filtered on SAINT score") + + geom_point(aes(x = log(NSAF), y = log2(FoldChange), size = SpecSum), colour = "black", shape = 21, data = a) + + labs(x = "ln(NSAF)") + if(label == TRUE & length(a$NSAF != 0)) { + p <- p + geom_text(data = a, aes(label = PreyGene, size = 10, vjust = 0, hjust = 0), colour = "black", show_guide = FALSE) + } + if(length(levels(data$Bait) > 1)) {p <- p + facet_wrap(~Bait, scales = "free_y")} + return(ggsave(p, width = 8, height = 4, filename = "bubble_zoom_NSAF.png")) + } +} +################################################################################################### +# Check Saintscore cutoff and stop program if not between 0 and 1 +################################################################################################### +cutoff_check <- function(cutoff){ + if( any(cutoff < 0 | cutoff > 1) ) stop('SAINT score cutoff not between 0 and 1. Please correct and try again') +} + +args <- commandArgs(trailingOnly = TRUE) +main(args[1], args[2], args[3], args[4], args[5], args[6], args[7], args[8], args[9])