Mercurial > repos > fcaramia > edger

comparison edgeR.pl @ 2:674c75219f15 draft

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Uploaded
author fcaramia
date Wed, 12 Sep 2012 23:44:45 -0400
parents
children e5fcbabbdea7
comparison
equal deleted inserted replaced
1:457a02a69f4d 2:674c75219f15
1 #/bin/perl
2
3 use strict;
4 use warnings;
5 use Getopt::Std;
6 use File::Basename;
7 use File::Path qw(make_path remove_tree);
8 $| = 1;
9
10 # Grab and set all options
11 my %OPTIONS = (a => "glm", d => "tag", f => "BH", p => 0.3, r => 5, u => "movingave");
12
13 getopts('a:d:e:f:h:lmn:o:p:r:tu:', \%OPTIONS);
14
15 die qq(
16 Usage: edgeR.pl [OPTIONS] factor::factor1::levels [factor::factor2::levels ...] cp::cont_pred1::values [cp::cont_pred2::values ...] cnt::contrast1 [cnt::contrast2] matrix
17
18 OPTIONS: -a STR Type Of Analysis [glm, pw, limma] (default: $OPTIONS{a})
19 -d STR The dispersion estimate to use for GLM analysis [tag, trend, common] (default: $OPTIONS{d})
20 -e STR Path to place additional output files
21 -f STR False discovery rate adjustment method [BH, holm, hochberg, hommel, BY, none] (default: $OPTIONS{f})
22 -h STR Name of html file for additional files
23 -l Output the normalised digital gene expression matrix in log2 format (only applicable when using limma and -n is also specified)
24 -m Perform all pairwise comparisons
25 -n STR File name to output the normalised digital gene expression matrix (only applicable when usinf glm or limma model)
26 -o STR File name to output csv file with results
27 -p FLT The proportion of all tags/genes to be used for the locally weighted estimation of the tagwise dispersion, ony applicable when 1 factor analysis selected (default: $OPTIONS{p})
28 -r INT Common Dispersion Rowsum Filter, ony applicable when 1 factor analysis selected (default: $OPTIONS{r})
29 -t Estimate Tagwise Disp when performing 1 factor analysis
30 -u STR Method for allowing the prior distribution for the dispersion to be abundance- dependent ["movingave", "tricube", "none"] (default: $OPTIONS{u})
31
32 ) if(!@ARGV);
33
34 my $matrix = pop @ARGV;
35
36 make_path($OPTIONS{e});
37 open(Rcmd,">$OPTIONS{e}/r_script.R") or die "Cannot open $OPTIONS{e}/r_script.R\n\n";
38 print Rcmd "
39 zz <- file(\"$OPTIONS{e}/r_script.err\", open=\"wt\")
40 sink(zz)
41 sink(zz, type=\"message\")
42
43 library(edgeR)
44 library(limma)
45
46 # read in matrix and groups
47 toc <- read.table(\"$matrix\", sep=\"\\t\", comment=\"\", as.is=T)
48 groups <- sapply(toc[1, -1], strsplit, \":\")
49 for(i in 1:length(groups)) { g <- make.names(groups[[i]][2]); names(groups)[i] <- g; groups[[i]] <- groups[[i]][-2] }
50 colnames(toc) <- make.names(toc[2,])
51 toc[,1] <- gsub(\",\", \".\", toc[,1])
52 tagnames <- toc[-(1:2), 1]
53 rownames(toc) <- toc[,1]
54 toc <- toc[-(1:2), -1]
55 for(i in colnames(toc)) toc[, i] <- as.numeric(toc[,i])
56 norm_factors <- calcNormFactors(as.matrix(toc))
57
58 pw_tests <- list()
59 uniq_groups <- unique(names(groups))
60 for(i in 1:(length(uniq_groups)-1)) for(j in (i+1):length(uniq_groups)) pw_tests[[length(pw_tests)+1]] <- c(uniq_groups[i], uniq_groups[j])
61 DGE <- DGEList(toc, lib.size=norm_factors*colSums(toc), group=names(groups))
62 pdf(\"$OPTIONS{e}/MA_plots_normalisation.pdf\", width=14)
63 for(i in 1:length(pw_tests)) {
64 j <- c(which(names(groups) == pw_tests[[i]][1])[1], which(names(groups) == pw_tests[[i]][2])[1])
65 par(mfrow = c(1, 2))
66 maPlot(toc[, j[1]], toc[, j[2]], normalize = TRUE, pch = 19, cex = 0.2, ylim = c(-10, 10), main=paste(\"MA Plot\", colnames(toc)[j[1]], \"vs\", colnames(toc)[j[2]]))
67 grid(col = \"blue\")
68 abline(h = log2(norm_factors[j[2]]), col = \"red\", lwd = 4)
69 maPlot(DGE\$counts[, j[1]]/DGE\$samples\$lib.size[j[1]], DGE\$counts[, j[2]]/DGE\$samples\$lib.size[j[2]], normalize = FALSE, pch = 19, cex = 0.2, ylim = c(-8, 8), main=paste(\"MA Plot\", colnames(toc)[j[1]], \"vs\", colnames(toc)[j[2]], \"Normalised\"))
70 grid(col = \"blue\")
71 }
72 dev.off()
73 pdf(file=\"$OPTIONS{e}/MDSplot.pdf\")
74 plotMDS(DGE, main=\"MDS Plot\", col=as.numeric(factor(names(groups)))+1, xlim=c(-3,3))
75 dev.off()
76 tested <- list()
77 ";
78
79 my $all_cont;
80 my @add_cont;
81 my @fact;
82 my @fact_names;
83 my @cp;
84 my @cp_names;
85 if(@ARGV) {
86 foreach my $input (@ARGV) {
87 my @tmp = split "::", $input;
88 if($tmp[0] eq "factor") {
89 $tmp[1] =~ s/[ \?\(\)\[\]\/\\=+<>:;\"\',\*\^\|\&-]/./g;
90 push @fact_names, $tmp[1];
91 $tmp[2] =~ s/:/\", \"/g;
92 $tmp[2] = "\"".$tmp[2]."\"";
93 push @fact, $tmp[2];
94 } elsif($tmp[0] eq "cp") {
95 $tmp[1] =~ s/[ \?\(\)\[\]\/\\=+<>:;\"\',\*\^\|\&-]/./g;
96 push @cp_names, $tmp[1];
97 $tmp[2] =~ s/:/, /g;
98 push @cp, $tmp[2];
99 } elsif($tmp[0] eq "cnt") {
100 push @add_cont, $tmp[1];
101 } else {
102 die("Unknown Input: $input\n");
103 }
104 }
105 }
106
107 if($OPTIONS{a} eq "pw") {
108 print Rcmd "
109 disp <- estimateCommonDisp(DGE, rowsum.filter=$OPTIONS{r})
110 ";
111 if(defined $OPTIONS{t}) {
112 print Rcmd "
113 disp <- estimateTagwiseDisp(disp, trend=\"$OPTIONS{u}\", prop.used=$OPTIONS{p})
114 pdf(file=\"$OPTIONS{e}/Tagwise_Dispersion_vs_Abundance.pdf\")
115 plot(log2(1e06*disp\$conc\$conc.common), disp\$tagwise.dispersion, xlab=\"Counts per million (log2 scale)\", ylab=\"Tagwise dispersion\")
116 abline(h=disp\$common.dispersion, col=\"firebrick\", lwd=3)
117 dev.off()
118 "
119 }
120 print Rcmd "
121 for(i in 1:length(pw_tests)) {
122 tested[[i]] <- exactTest(disp, pair=pw_tests[[i]])
123 names(tested)[i] <- paste(pw_tests[[i]][2], \"-\", pw_tests[[i]][1], sep=\"\")
124 }
125 pdf(file=\"$OPTIONS{e}/Smear_Plots.pdf\")
126 for(i in 1:length(pw_tests)) {
127 if(nrow(decideTestsDGE(tested[[i]] , p.value=0.05)) > 0) {
128 de_tags <- rownames(decideTestsDGE(tested[[i]] , p.value=0.05, adjust.method=\"$OPTIONS{f}\"))
129 ttl <- \"(Diff. Exp. Genes With adj. Pvalue < 0.05 highlighted)\"
130 } else {
131 de_tags <- rownames(topTags(tested[[i]], n=100)\$table)
132 ttl <- \"(Top 100 tags highlighted)\"
133 }
134
135 plotSmear(disp, pair=pw_tests[[i]], de.tags = de_tags, main = paste(\"FC plot\", ttl))
136 abline(h = c(-2, 2), col = \"dodgerblue\")
137 }
138 dev.off()
139 ";
140 } elsif($OPTIONS{a} eq "glm") {
141 for(my $fct = 0; $fct <= $#fact_names; $fct++) {
142 print Rcmd "
143 $fact_names[$fct] <- c($fact[$fct])
144 ";
145 }
146 for(my $fct = 0; $fct <= $#cp_names; $fct++) {
147 print Rcmd "
148 $cp_names[$fct] <- c($cp[$fct])
149 ";
150 }
151 my $all_fact = "";
152 if(@fact_names) {
153 foreach (@fact_names) {
154 $all_fact .= " + factor($_)";
155 }
156 }
157 my $all_cp = "";
158 if(@cp_names) {
159 $all_cp = " + ".join(" + ", @cp_names);
160 }
161 print Rcmd "
162 group_fact <- factor(names(groups))
163 design <- model.matrix(~ -1 + group_fact${all_fact}${all_cp})
164 colnames(design) <- sub(\"group_fact\", \"\", colnames(design))
165 ";
166 foreach my $fct (@fact_names) {
167 print Rcmd "
168 colnames(design) <- make.names(sub(\"factor.$fct.\", \"\", colnames(design)))
169 ";
170 }
171 print Rcmd "
172 disp <- estimateGLMCommonDisp(DGE, design)
173 ";
174 if($OPTIONS{d} eq "tag" || $OPTIONS{d} eq "trend") {
175 print Rcmd "
176 disp <- estimateGLMTrendedDisp(disp, design)
177 ";
178 }
179 if($OPTIONS{d} eq "tag") {
180 print Rcmd "
181 disp <- estimateGLMTagwiseDisp(disp, design)
182 fit <- glmFit(disp, design)
183 pdf(file=\"$OPTIONS{e}/Tagwise_Dispersion_vs_Abundance.pdf\")
184 plot(fit\$abund+log(1e06), sqrt(disp\$tagwise.dispersion), xlab=\"Counts per million (log2 scale)\", ylab=\"Tagwise dispersion\")
185 oo <- order(disp\$abundance)
186 lines(fit\$abundance[oo]+log(1e06), sqrt(disp\$trended.dispersion[oo]), col=\"dodgerblue\", lwd=3)
187 abline(h=sqrt(disp\$common.dispersion), col=\"firebrick\", lwd=3)
188 dev.off()
189 ";
190 }
191 if(@add_cont) {
192 $all_cont = "\"".join("\", \"", @add_cont)."\"";
193 print Rcmd "
194 cont <- c(${all_cont})
195 for(i in uniq_groups) cont <- gsub(paste(groups[[i]], \"([^0-9])\", sep=\"\"), paste(i, \"\\\\1\", sep=\"\"), cont)
196 for(i in uniq_groups) cont <- gsub(paste(groups[[i]], \"\$\", sep=\"\"), i, cont)
197 ";
198 } else {
199 print Rcmd "
200 cont <- NULL
201 ";
202 }
203 if(defined $OPTIONS{m}) {
204 print Rcmd "
205 for(i in 1:length(pw_tests)) cont <- c(cont, paste(pw_tests[[i]][2], \"-\", pw_tests[[i]][1], sep=\"\"))
206 ";
207 }
208 if(!defined $OPTIONS{m} && !@add_cont){
209 die("No Contrasts have been specified, you must at least either select multiple pairwise comparisons or specify a custom contrast\n");
210 }
211 print Rcmd "
212 fit <- glmFit(disp, design)
213 cont <- makeContrasts(contrasts=cont, levels=design)
214 for(i in colnames(cont)) tested[[i]] <- glmLRT(disp, fit, contrast=cont[,i])
215 ";
216 if(defined $OPTIONS{n}) {
217 print Rcmd "
218 tab <- data.frame(ID=rownames(fit\$fitted.values), fit\$fitted.values, stringsAsFactors=F)
219 write.table(tab, \"$OPTIONS{n}\", quote=F, sep=\"\\t\", row.names=F)
220 ";
221 }
222 } elsif($OPTIONS{a} eq "limma") {
223 for(my $fct = 0; $fct <= $#fact_names; $fct++) {
224 print Rcmd "
225 $fact_names[$fct] <- c($fact[$fct])
226 ";
227 }
228 for(my $fct = 0; $fct <= $#cp_names; $fct++) {
229 print Rcmd "
230 $cp_names[$fct] <- c($cp[$fct])
231 ";
232 }
233 my $all_fact = "";
234 if(@fact_names) {
235 foreach (@fact_names) {
236 $all_fact .= " + factor($_)";
237 }
238 }
239 my $all_cp = "";
240 if(@cp_names) {
241 $all_cp = " + ".join(" + ", @cp_names);
242 }
243 print Rcmd "
244 group_fact <- factor(names(groups))
245 design <- model.matrix(~ -1 + group_fact${all_fact}${all_cp})
246 colnames(design) <- sub(\"group_fact\", \"\", colnames(design))
247 ";
248 foreach my $fct (@fact_names) {
249 print Rcmd "
250 colnames(design) <- make.names(sub(\"factor.$fct.\", \"\", colnames(design)))
251 ";
252 }
253 print Rcmd "
254 isexpr <- rowSums(cpm(toc)>1) >= 2
255 toc <- toc[isexpr, ]
256 pdf(file=\"$OPTIONS{e}/LIMMA_voom.pdf\")
257 y <- voom(toc, design, plot=TRUE, lib.size=colSums(toc)*norm_factors)
258 dev.off()
259
260 pdf(file=\"$OPTIONS{e}/LIMMA_MDS_plot.pdf\")
261 plotMDS(y, labels=colnames(toc), col=as.numeric(factor(names(groups)))+1, gene.selection=\"common\")
262 dev.off()
263 fit <- lmFit(y, design)
264 ";
265 if(defined $OPTIONS{n}) {
266 if(defined $OPTIONS{l}) {
267 print Rcmd "
268 tab <- data.frame(ID=rownames(y\$E), y\$E, stringsAsFactors=F)
269 ";
270 } else {
271 print Rcmd "
272 tab <- data.frame(ID=rownames(y\$E), 2^y\$E, stringsAsFactors=F)
273 ";
274 }
275 print Rcmd "
276 write.table(tab, \"$OPTIONS{n}\", quote=F, sep=\"\\t\", row.names=F)
277 ";
278 }
279 if(@add_cont) {
280 $all_cont = "\"".join("\", \"", @add_cont)."\"";
281 print Rcmd "
282 cont <- c(${all_cont})
283 for(i in uniq_groups) cont <- gsub(paste(groups[[i]], \"([^0-9])\", sep=\"\"), paste(i, \"\\\\1\", sep=\"\"), cont)
284 for(i in uniq_groups) cont <- gsub(paste(groups[[i]], \"\$\", sep=\"\"), i, cont)
285 ";
286 } else {
287 print Rcmd "
288 cont <- NULL
289 ";
290 }
291 if(defined $OPTIONS{m}) {
292 print Rcmd "
293 for(i in 1:length(pw_tests)) cont <- c(cont, paste(pw_tests[[i]][2], \"-\", pw_tests[[i]][1], sep=\"\"))
294 ";
295 }
296 if(!defined $OPTIONS{m} && !@add_cont){
297 die("No Contrasts have been specified, you must at least either select multiple pairwise comparisons or specify a custom contrast\n");
298 }
299 print Rcmd "
300 cont <- makeContrasts(contrasts=cont, levels=design)
301 fit2 <- contrasts.fit(fit, cont)
302 fit2 <- eBayes(fit2)
303 ";
304 } else {
305 die("Anaysis type $OPTIONS{a} not found\n");
306
307 }
308
309 if($OPTIONS{a} ne "limma") {
310 print Rcmd "
311 options(digits = 6)
312 tab <- NULL
313 for(i in names(tested)) {
314 tab_tmp <- topTags(tested[[i]], n=Inf, adjust.method=\"$OPTIONS{f}\")[[1]]
315 colnames(tab_tmp) <- paste(i, colnames(tab_tmp), sep=\":\")
316 tab_tmp <- tab_tmp[tagnames,]
317 if(is.null(tab)) {
318 tab <- tab_tmp
319 } else tab <- cbind(tab, tab_tmp)
320 }
321 tab <- cbind(Feature=rownames(tab), tab)
322 ";
323 } else {
324 print Rcmd "
325 tab <- NULL
326 options(digits = 6)
327 for(i in colnames(fit2)) {
328 tab_tmp <- topTable(fit2, coef=i, n=Inf, sort.by=\"none\", adjust.method=\"$OPTIONS{f}\")
329 colnames(tab_tmp)[-1] <- paste(i, colnames(tab_tmp)[-1], sep=\":\")
330 if(is.null(tab)) {
331 tab <- tab_tmp
332 } else tab <- cbind(tab, tab_tmp[,-1])
333 }
334 ";
335 }
336 print Rcmd "
337 write.table(tab, \"$OPTIONS{o}\", quote=F, sep=\"\\t\", row.names=F)
338 sink(type=\"message\")
339 sink()
340 ";
341 close(Rcmd);
342 system("R --no-restore --no-save --no-readline < $OPTIONS{e}/r_script.R > $OPTIONS{e}/r_script.out");
343
344 open(HTML, ">$OPTIONS{h}");
345 print HTML "<html><head><title>EdgeR: Empirical analysis of digital gene expression data</title></head><body><h3>EdgeR Additional Files:</h3><p><ul>\n";
346 print HTML "<li><a href=MA_plots_normalisation.pdf>MA_plots_normalisation.pdf</a></li>\n";
347 print HTML "<li><a href=MDSplot.pdf>MDSplot.pdf</a></li>\n";
348 if($OPTIONS{a} eq "pw") {
349 if(defined $OPTIONS{t}) {
350 print HTML "<li><a href=Tagwise_Dispersion_vs_Abundance.pdf>Tagwise_Dispersion_vs_Abundance.pdf</a></li>\n";
351 }
352 print HTML "<li><a href=Smear_Plots.pdf>Smear_Plots.pdf</a></li>\n";
353 } elsif($OPTIONS{a} eq "glm" && $OPTIONS{d} eq "tag") {
354 print HTML "<li><a href=Tagwise_Dispersion_vs_Abundance.pdf>Tagwise_Dispersion_vs_Abundance.pdf</a></li>\n";
355 } elsif($OPTIONS{a} eq "limma") {
356 print HTML "<li><a href=LIMMA_MDS_plot.pdf>LIMMA_MDS_plot.pdf</a></li>\n";
357 print HTML "<li><a href=LIMMA_voom.pdf>LIMMA_voom.pdf</a></li>\n";
358 }
359 print HTML "<li><a href=r_script.R>r_script.R</a></li>\n";
360 print HTML "<li><a href=r_script.out>r_script.out</a></li>\n";
361 print HTML "<li><a href=r_script.err>r_script.err</a></li>\n";
362 print HTML "</ul></p>\n";
363 close(HTML);
364