comparison Rscripts/ridb-regression.R @ 0:9d5f4f5f764b

Initial commit to toolshed
author pieter.lukasse@wur.nl
date Thu, 16 Jan 2014 13:10:00 +0100
parents
children
comparison
equal deleted inserted replaced
-1:000000000000 0:9d5f4f5f764b
1 ##
2 #
3 # Performs regression analysis using either 3rd degree polynomial- or linear-method
4 #
5 ##
6
7 # Commandline arguments
8 args <- commandArgs(TRUE)
9 if (length(args) < 7)
10 stop(cat("Missing arguments, usage:\n\tRscript ridb-regression.R RI-database ",
11 "ouput_file logfile min_residuals range_mod pvalue rsquared method ",
12 "plot(yes/no) plot_archive"))
13
14 ridb <- args[1]
15 out_file <- args[2]
16 logfile <- args[3]
17 min_residuals <- as.integer(args[4])
18 range_mod <- as.integer(args[5])
19 pvalue <- as.double(args[6])
20 rsquared <- as.double(args[7])
21 method <- args[8]
22 plot <- tolower(args[9])
23 if (plot == 'true')
24 plot_archive = args[10]
25
26 # Do not show warnings etc.
27 sink(file='/dev/null')
28
29 progress <- c()
30 logger <- function(logdata) {
31 ## Logs progress, adds a timestamp for each event
32 #cat(paste(Sys.time(), "\t", logdata, "\n", sep="")) ## DEBUG
33 progress <<- c(progress, paste(Sys.time(), "\t", logdata, sep=""))
34 }
35
36 logger("Reading Retention Index Database..")
37
38 # Read Retention Index Database
39 ridb <- read.csv(ridb, header=TRUE, sep="\t")
40 logger(paste("\t", nrow(ridb), "records read.."))
41 # Get a unique list
42 gc_columns <- unique(as.vector(as.matrix(ridb['Column.name'])[,1]))
43 cas_numbers <- unique(as.vector(as.matrix(ridb['CAS'])[,1]))
44
45 add_poly_fit <- function(fit, gc1_index, gc2_index, range) {
46 pval = anova.lm(fit)$Pr
47 r.squared = summary(fit)$r.squared
48
49 data = rep(NA, 11)
50 # Append results to matrix
51 data[1] = gc_columns[gc1_index] # Column 1
52 data[2] = gc_columns[gc2_index] # Column 2
53 data[3] = coefficients(fit)[1] # The 4 coefficients
54 data[4] = coefficients(fit)[2]
55 data[5] = coefficients(fit)[3]
56 data[6] = coefficients(fit)[4]
57 data[7] = range[1] # Left limit
58 data[8] = range[2] # Right limit
59 data[9] = length(fit$residuals) # Number of datapoints analysed
60 data[10] = pval[1] # p-value for resulting fitting
61 data[11] = r.squared # R-squared
62 return(data)
63 }
64
65
66 add_linear_fit <- function(fit, gc1_index, gc2_index, range) {
67 pval = anova.lm(fit)$Pr
68 r.squared = summary(fit)$r.squared
69
70 data = rep(NA, 7)
71 # Append results to matrix
72 data[1] = gc_columns[gc1_index] # Column 1
73 data[2] = gc_columns[gc2_index] # Column 2
74 data[3] = coefficients(fit)[1] # The 4 coefficients
75 data[4] = coefficients(fit)[2]
76 data[7] = length(fit$residuals) # Number of datapoints analysed
77 data[8] = pval[1] # p-value for resulting fitting
78 data[9] = r.squared # R-squared
79 return(data)
80 }
81
82
83 add_fit <- function(fit, gc1_index, gc2_index, range, method) {
84 if (method == 'poly')
85 return(add_poly_fit(fit, gc1_index, gc2_index, range))
86 else
87 return(add_linear_fit(fit, gc1_index, gc2_index, range))
88 }
89
90
91 plot_fit <- function(ri1, ri2, gc1_index, gc2_index, coeff, range, method) {
92 if (method == 'poly')
93 pol <- function(x) coeff[4]*x^3 + coeff[3]*x^2 + coeff[2]*x + coeff[1]
94 else
95 pol <- function(x) coeff[2]*x + coeff[1]
96 pdf(paste('regression_model_',
97 make.names(gc_columns[gc1_index]), '_vs_',
98 make.names(gc_columns[gc2_index]), '.pdf', sep=''))
99 curve(pol, 250:3750, col="red", lwd=2.5, main='Regression Model', xlab=gc_columns[gc1_index],
100 ylab=gc_columns[gc2_index], xlim=c(250, 3750), ylim=c(250, 3750))
101 points(ri1, ri2, lwd=0.4)
102 # Add vertical lines showing left- and right limits when using poly method
103 if (method == 'poly')
104 abline(v=range, col="grey", lwd=1.5)
105 dev.off()
106 }
107
108 # Initialize output dataframe
109 if (method == 'poly') {
110 m <- data.frame(matrix(ncol = 11, nrow = 10))
111 } else {
112 m <- data.frame(matrix(ncol = 9, nrow = 10))
113 }
114
115
116 get_fit <- function(gc1, gc2, method) {
117 if (method == 'poly')
118 return(lm(gc1 ~ poly(gc2, 3, raw=TRUE)))
119 else
120 return(lm(gc1 ~ gc2))
121 }
122
123 # Permutate
124 k <- 1
125 logger(paste("Permutating (with ", length(gc_columns), " GC-columns)..", sep=""))
126
127 for (i in 1:(length(gc_columns)-1)) {
128 logger(paste("\tCalculating model for ", gc_columns[i], "..", sep=""))
129 breaks <- 0
130 for (j in (i+1):length(gc_columns)) {
131 col1 = ridb[which(ridb['Column.name'][,1] == gc_columns[i]),]
132 col2 = ridb[which(ridb['Column.name'][,1] == gc_columns[j]),]
133
134 # Find CAS numbers for which both columns have data (intersect)
135 cas_intersect = intersect(col1[['CAS']], col2[['CAS']])
136
137 # Skip if number of shared CAS entries is < cutoff
138 if (length(cas_intersect) < min_residuals) {
139 breaks = breaks + 1
140 next
141 }
142 # Gather Retention Indices
143 col1_data = col1[['RI']][match(cas_intersect, col1[['CAS']])]
144 col2_data = col2[['RI']][match(cas_intersect, col2[['CAS']])]
145
146 # Calculate the range within which regression is possible (and move if 'range_mod' != 0)
147 range = c(min(c(min(col1_data), min(col2_data))), max(c(max(col1_data), max(col2_data))))
148 if (range_mod != 0) {
149 # Calculate percentage and add/subtract from range
150 perc = diff(range) / 100
151 perc_cutoff = range_mod * perc
152 range = as.integer(range + c(perc_cutoff, -perc_cutoff))
153 }
154
155 # Calculate model for column1 vs column2 and plot if requested
156 fit = get_fit(col1_data, col2_data, method)
157 m[k,] = add_fit(fit, i, j, range, method)
158
159 if (plot == 'true')
160 plot_fit(col1_data, col2_data, i, j, coefficients(fit), range, method)
161
162 # Calculate model for column2 vs column1 and plot if requested
163 fit = get_fit(col2_data, col1_data, method)
164 m[k + 1,] = add_fit(fit, j, i, range, method)
165
166 if (plot == 'true')
167 plot_fit(col2_data, col1_data, j, i, coefficients(fit), range, method)
168
169 k = k + 2
170 }
171 logger(paste("\t\t", breaks, " comparisons have been skipped due to nr. of datapoints < cutoff", sep=""))
172 }
173
174 # Filter on pvalue and R-squared
175 logger("Filtering on pvalue and R-squared..")
176 if (method == 'poly') {
177 pval_index <- which(m[,10] < pvalue)
178 rsquared_index <- which(m[,11] > rsquared)
179 } else {
180 pval_index <- which(m[,8] < pvalue)
181 rsquared_index <- which(m[,9] > rsquared)
182 }
183 logger(paste(nrow(m) - length(pval_index), " models discarded due to pvalue > ", pvalue, sep=""))
184
185 logger(paste(nrow(m) - length(rsquared_index), " models discarded due to R-squared < ", rsquared, sep=""))
186
187 # Remaining rows
188 index = unique(c(pval_index, rsquared_index))
189
190 # Reduce dataset
191 m = m[index,]
192 sink()
193
194 # Place plots in the history as a ZIP file
195 if (plot == 'true') {
196 logger("Creating archive with model graphics..")
197 system(paste("zip -9 -r models.zip *.pdf > /dev/null", sep=""))
198 system(paste("cp models.zip ", plot_archive, sep=""))
199 }
200
201 # Save dataframe as tab separated file
202 logger("All done, saving data..")
203 header = c("Column1", "Column2", "Coefficient1", "Coefficient2", "Coefficient3", "Coefficient4",
204 "LeftLimit", "RightLimit", "Residuals", "pvalue", "Rsquared")
205 if (method != 'poly')
206 header = header[c(1:4, 7:11)]
207 write(progress, logfile)
208 write.table(m, file=out_file, sep="\t", quote=FALSE, col.names=header, row.names=FALSE)