diff lda_analy.xml @ 0:f38763b52f33 draft

Imported from capsule None
author devteam
date Mon, 28 Jul 2014 11:56:39 -0400
parents
children f0b6217f4a0c
line wrap: on
line diff
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/lda_analy.xml	Mon Jul 28 11:56:39 2014 -0400
@@ -0,0 +1,288 @@
+<tool id="lda_analy1" name="Perform LDA" version="1.0.1">
+	<description>Linear Discriminant Analysis</description>
+    <requirements>
+      <requirement type="package" version="2.11.0">R</requirement>
+    </requirements>
+	<command interpreter="sh">r_wrapper.sh $script_file</command>
+	<inputs>
+		<param format="tabular" name="input" type="data" label="Source file"/>
+		<param name="cond" size="30" type="integer" value="3" label="Number of principal components" help="See TIP below">
+			<validator type="empty_field" message="Enter a valid number of principal components, see syntax below for examples"/>
+		</param>
+
+	</inputs>
+	<outputs>
+		<data format="txt" name="output" />
+	</outputs>
+
+	<tests>
+		<test>
+			<param name="input" value="matrix_generator_for_pc_and_lda_output.tabular"/>
+			<output name="output" file="lda_analy_output.txt"/>
+			<param name="cond" value="2"/>
+
+		</test>
+	</tests>
+
+	<configfiles>
+        	<configfile name="script_file">
+
+        rm(list = objects() )
+
+        ############# FORMAT X DATA #########################
+        format&lt;-function(data) {
+            ind=NULL
+            for(i in 1 : ncol(data)){
+                if (is.na(data[nrow(data),i])) {
+                    ind&lt;-c(ind,i)
+                }
+            }
+            #print(is.null(ind))
+            if (!is.null(ind)) {
+                data&lt;-data[,-c(ind)]
+            }
+
+            data
+        }
+
+        ########GET RESPONSES ###############################
+        get_resp&lt;- function(data) {
+            resp1&lt;-as.vector(data[,ncol(data)])
+                resp=numeric(length(resp1))
+            for (i in 1:length(resp1)) {
+                if (resp1[i]=="Y ") {
+                    resp[i] = 0
+                }
+                if (resp1[i]=="X ") {
+                    resp[i] = 1
+                }
+            }
+                return(resp)
+        }
+
+        ######## CHARS TO NUMBERS ###########################
+        f_to_numbers&lt;- function(F) { 
+            ind&lt;-NULL
+            G&lt;-matrix(0,nrow(F), ncol(F))
+            for (i in 1:nrow(F)) {
+                for (j in 1:ncol(F)) {
+                    G[i,j]&lt;-as.integer(F[i,j])
+                }
+            }
+            return(G)
+        }
+
+        ###################NORMALIZING#########################
+        norm &lt;- function(M, a=NULL, b=NULL) {
+            C&lt;-NULL
+            ind&lt;-NULL
+
+            for (i in 1: ncol(M)) {
+                if (sd(M[,i])!=0) {
+                    M[,i]&lt;-(M[,i]-mean(M[,i]))/sd(M[,i])
+                }
+                #   else {print(mean(M[,i]))}   
+            }
+            return(M)
+        }
+
+        ##### LDA DIRECTIONS #################################
+        lda_dec &lt;- function(data, k){
+            priors=numeric(k)
+            grandmean&lt;-numeric(ncol(data)-1)
+            means=matrix(0,k,ncol(data)-1)
+            B = matrix(0, ncol(data)-1, ncol(data)-1)
+            N=nrow(data)
+            for (i in 1:k){
+                priors[i]=sum(data[,1]==i)/N
+                grp=subset(data,data\$group==i)
+                means[i,]=mean(grp[,2:ncol(data)])
+                #print(means[i,])
+                #print(priors[i])
+                #print(priors[i]*means[i,])
+                grandmean = priors[i]*means[i,] + grandmean           
+            }
+
+            for (i in 1:k) {
+                B= B + priors[i]*((means[i,]-grandmean)%*%t(means[i,]-grandmean))
+            }
+    
+            W = var(data[,2:ncol(data)])
+            svdW = svd(W)
+            inv_sqrtW =solve(svdW\$v %*% diag(sqrt(svdW\$d)) %*% t(svdW\$v))
+            B_star= t(inv_sqrtW)%*%B%*%inv_sqrtW
+            B_star_decomp = svd(B_star)
+            directions  = inv_sqrtW%*%B_star_decomp\$v
+            return( list(directions, B_star_decomp\$d) )                          
+        }
+
+        ################ NAIVE BAYES FOR 1D SIR OR LDA ##############
+        naive_bayes_classifier &lt;- function(resp, tr_data, test_data, k=2, tau) {
+            tr_data=data.frame(resp=resp, dir=tr_data)
+            means=numeric(k)
+            #print(k)
+            cl=numeric(k)
+            predclass=numeric(length(test_data))
+            for (i in 1:k) {
+                grp = subset(tr_data, resp==i)
+                means[i] = mean(grp\$dir)
+            #print(i, means[i])  
+            }
+            cutoff = tau*means[1]+(1-tau)*means[2] 
+            #print(tau)
+            #print(means)
+            #print(cutoff)
+            if (cutoff&gt;means[1]) {
+               cl[1]=1 
+               cl[2]=2
+            }
+            else {
+               cl[1]=2 
+               cl[2]=1
+            }
+
+            for (i in 1:length(test_data)) {
+
+                if (test_data[i] &lt;= cutoff) {
+                    predclass[i] = cl[1]
+            }
+                else {
+                    predclass[i] = cl[2] 
+            }  
+                }
+            #print(means)
+            #print(mean(means))
+            #X11()
+            #plot(test_data,pch=predclass, col=resp) 
+            predclass
+        }
+
+        ################# EXTENDED ERROR RATES #################
+        ext_error_rate &lt;- function(predclass, actualclass,msg=c("you forgot the message"), pr=1) {
+                 er=sum(predclass != actualclass)/length(predclass)
+
+                 matr&lt;-data.frame(predclass=predclass,actualclass=actualclass)
+                 escapes = subset(matr, actualclass==1)
+                 subjects = subset(matr, actualclass==2)      
+                 er_esc=sum(escapes\$predclass != escapes\$actualclass)/length(escapes\$predclass) 
+                 er_subj=sum(subjects\$predclass != subjects\$actualclass)/length(subjects\$predclass)   
+
+                 if (pr==1) {
+        #             print(paste(c(msg, 'overall : ', (1-er)*100, "%."),collapse=" "))
+        #             print(paste(c(msg, 'within escapes : ', (1-er_esc)*100, "%."),collapse=" "))
+        #             print(paste(c(msg, 'within subjects: ', (1-er_subj)*100, "%."),collapse=" ")) 
+            }
+            return(c((1-er)*100, (1-er_esc)*100, (1-er_subj)*100))                                                                                    
+        }
+
+        ## Main Function ##
+
+	files&lt;-matrix("${input}", 1,1, byrow=T)
+
+	d&lt;-"${cond}"   # Number of PC
+
+	tau&lt;-seq(0,1, by=0.005)
+	#tau&lt;-seq(0,1, by=0.1)
+	for_curve=matrix(-10, 3,length(tau))
+
+	##############################################################
+
+	test_data_whole_X &lt;-read.delim(files[1,1], row.names=1)
+
+	#### FORMAT TRAINING DATA ####################################
+	# get only necessary columns 
+
+	test_data_whole_X&lt;-format(test_data_whole_X)
+	oligo_labels&lt;-test_data_whole_X[1:(nrow(test_data_whole_X)-1),ncol(test_data_whole_X)]
+	test_data_whole_X&lt;-test_data_whole_X[,1:(ncol(test_data_whole_X)-1)]
+
+	X_names&lt;-colnames(test_data_whole_X)[1:ncol(test_data_whole_X)]
+	test_data_whole_X&lt;-t(test_data_whole_X)
+	resp&lt;-get_resp(test_data_whole_X) 
+	ldaqda_resp = resp + 1
+	a&lt;-sum(resp)		# Number of Subject
+	b&lt;-length(resp) - a	# Number of Escape   
+	## FREQUENCIES #################################################
+	F&lt;-test_data_whole_X[,1:(ncol(test_data_whole_X)-1)]
+	F&lt;-f_to_numbers(F)
+	FN&lt;-norm(F, a, b)
+	ss&lt;-svd(FN)
+	eigvar&lt;-NULL
+	eig&lt;-ss\$d^2
+
+	for ( i in 1:length(ss\$d)) {
+		eigvar[i]&lt;-sum(eig[1:i])/sum(eig)
+	}
+
+	#print(paste(c("Variance explained : ", eigvar[d]*100, "%"), collapse=""))
+	
+	Z&lt;-F%*%ss\$v
+
+	ldaqda_data &lt;- data.frame(group=ldaqda_resp,Z[,1:d])
+	lda_dir&lt;-lda_dec(ldaqda_data,2)
+	train_lda_pred &lt;-Z[,1:d]%*%lda_dir[[1]]
+
+	############# NAIVE BAYES CROSS-VALIDATION #############
+	### LDA #####
+
+	y&lt;-ldaqda_resp
+	X&lt;-F
+	cv&lt;-matrix(c(rep('NA',nrow(test_data_whole_X))), nrow(test_data_whole_X), length(tau))
+	for (i in 1:nrow(test_data_whole_X)) {
+	#	print(i)
+		resp&lt;-y[-i]
+		p&lt;-matrix(X[-i,], dim(X)[1]-1, dim(X)[2])
+		testdata&lt;-matrix(X[i,],1,dim(X)[2])
+		p1&lt;-norm(p)
+		sss&lt;-svd(p1)
+		pred&lt;-(p%*%sss\$v)[,1:d]
+		test&lt;- (testdata%*%sss\$v)[,1:d]
+		lda  &lt;- lda_dec(data.frame(group=resp,pred),2)
+		pred &lt;- pred[,1:d]%*%lda[[1]][,1]
+		test &lt;- test%*%lda[[1]][,1]
+		test&lt;-matrix(test, 1, length(test))
+		for (t in 1:length(tau)) {
+			cv[i, t] &lt;- naive_bayes_classifier (resp, pred, test,k=2, tau[t]) 
+		}
+ 	}
+
+	for (t in 1:length(tau)) {
+		tr_err&lt;-ext_error_rate(cv[,t], ldaqda_resp , c("CV"), 1)
+		for_curve[1:3,t]&lt;-tr_err
+	}
+
+	dput(for_curve, file="${output}")
+
+
+		</configfile>
+	</configfiles>
+
+	<help>
+
+.. class:: infomark
+
+**TIP:** If you want to perform Principal Component Analysis (PCA) on the give numeric input data (which corresponds to the "Source file First in "Generate A Matrix" tool), please use *Multivariate Analysis/Principal Component Analysis*
+
+-----
+
+.. class:: infomark
+
+**What it does**
+
+This tool consists of the module to perform the Linear Discriminant Analysis as described in Carrel et al., 2006 (PMID: 17009873)
+
+*Carrel L, Park C, Tyekucheva S, Dunn J, Chiaromonte F, et al. (2006) Genomic Environment Predicts Expression Patterns on the Human 	Inactive X Chromosome. PLoS Genet 2(9): e151. doi:10.1371/journal.pgen.0020151*
+
+-----
+
+.. class:: warningmark
+
+**Note**
+
+- Output from "Generate A Matrix" tool is used as input file for this tool 
+- Output of this tool contains LDA classification success rates for different values of the turning parameter tau (from 0 to 1 with 0.005 interval). This output file will be used to establish the ROC plot, and you can obtain more detail information from this plot. 
+
+
+</help>
+
+</tool>