Mercurial > repos > fabio > iwtomics
comparison loadandplot.R @ 83:f81d72e482cf draft
Uploaded
author | fabio |
---|---|
date | Fri, 06 Oct 2017 11:06:44 -0400 |
parents | |
children |
comparison
equal
deleted
inserted
replaced
82:8e397b62df4c | 83:f81d72e482cf |
---|---|
1 if (require("IWTomics",character.only = TRUE,quietly = FALSE)) { | |
2 args=commandArgs(TRUE) | |
3 | |
4 # get args names and values | |
5 args_values=strsplit(args,'=') | |
6 args_names=unlist(lapply(args_values,function(arg) arg[1])) | |
7 names(args_values)=args_names | |
8 args_values=lapply(args_values,function(arg) arg[2]) | |
9 # read filenames | |
10 outrdata=args_values$outrdata | |
11 outregions=args_values$outregions | |
12 outfeatures=args_values$outfeatures | |
13 outpdf=args_values$outpdf | |
14 regionspaths=unlist(strsplit(args_values$regionspaths,'\\|')) | |
15 if("regionsheaderfile" %in% args_names){ | |
16 # the file regionsheaderfile must contain as first column the (unique) regionsfilenames, | |
17 # as second column the corresponding ids and as third column the names | |
18 tryCatch({ | |
19 regionsheader=read.delim(args_values$regionsheaderfile,header=FALSE,stringsAsFactors=FALSE,row.names=1,sep="\t") | |
20 regionsfilenames=unlist(strsplit(args_values$regionsfilenames,'\\|')) | |
21 if(length(setdiff(regionsfilenames,row.names(regionsheader)))) { | |
22 write("IWTomics message: Not all region files are present in the first column of header file for regions.", stderr()) | |
23 quit(save="no", status=11) | |
24 } | |
25 id_regions=regionsheader[regionsfilenames,1] | |
26 name_regions=regionsheader[regionsfilenames,2] | |
27 }, error = function(err) { | |
28 write("IWTomics message: An error has occurred reading the header file for regions. Please try again.", stderr()) | |
29 quit(save="no", status=10) #error on header file | |
30 }) | |
31 }else{ | |
32 eval(parse(text=args[[which(args_names=='regionsgalaxyids')]])) | |
33 id_regions=paste0('data_',regionsgalaxyids) | |
34 name_regions=paste0('data_',regionsgalaxyids) | |
35 } | |
36 featurespaths=unlist(strsplit(args_values$featurespaths,'\\|')) | |
37 if("featuresheaderfile" %in% args_names){ | |
38 # the file featuresheaderfile must contain as first column the (unique) featuresfilenames, | |
39 # as second column the corresponding ids and as third column the names | |
40 tryCatch({ | |
41 featuresheader=read.delim(args_values$featuresheaderfile,header=FALSE,stringsAsFactors=FALSE,row.names=1,sep="\t") | |
42 featuresfilenames=unlist(strsplit(args_values$featuresfilenames,'\\|')) | |
43 if(length(setdiff(featuresfilenames,row.names(featuresheader)))) { | |
44 write("IWTomics message: Not all feature files are present in the first column of header file for features.", stderr()) | |
45 quit(save="no", status=21) | |
46 } | |
47 id_features=featuresheader[featuresfilenames,1] | |
48 name_features=featuresheader[featuresfilenames,2] | |
49 }, error = function(err) { | |
50 write("IWTomics message: An error has occurred reading the header file for features. Please try again.", stderr()) | |
51 quit(save="no", status=20) #error on header file | |
52 }) | |
53 }else{ | |
54 eval(parse(text=args[[which(args_names=='featuresgalaxyids')]])) | |
55 id_features=paste0('data_',featuresgalaxyids) | |
56 name_features=paste0('data_',featuresgalaxyids) | |
57 } | |
58 # read parameters (from smoothing on) | |
59 i_smoothing=which(args_names=='smoothing') | |
60 for(i in i_smoothing:length(args)){ | |
61 eval(parse(text=args[[i]])) | |
62 } | |
63 | |
64 # load data | |
65 tryCatch({ | |
66 regionsFeatures=IWTomicsData(regionspaths,featurespaths,alignment, | |
67 id_regions,name_regions,id_features,name_features,start.are.0based=start.are.0based) | |
68 }, error = function(err) { | |
69 if(grepl('invalid format',err$message)){ | |
70 write("IWTomics message: Not enough columns in input file.", stderr()) | |
71 quit(save="no", status=31) # error, not enough columns in input file | |
72 }else if(grepl('duplicated regions',err$message)){ | |
73 write("IWTomics message: Duplicated regions in region file.", stderr()) | |
74 quit(save="no", status=32) # error, duplicated regions in region file | |
75 }else if(grepl('duplicated windows',err$message)){ | |
76 write("IWTomics message: Duplicated windows in feature file.", stderr()) | |
77 quit(save="no", status=33) # error, duplicated windows in feature file | |
78 }else if(grepl('overlapping windows',err$message)){ | |
79 write("IWTomics message: Overlapping windows in feature file.", stderr()) | |
80 quit(save="no", status=34) # error, overlapping windows in feature file | |
81 }else if(grepl('not all regions in datasets',err$message)){ | |
82 write("IWTomics message: Windows in feature files do not cover all regions in region files.", stderr()) | |
83 quit(save="no", status=35) # error, windows in feature files do not cover all regions in region files | |
84 }else if(grepl('ifferent size windows',err$message)){ | |
85 write("IWTomics message: All windows in a feature file must have the same size.", stderr()) | |
86 quit(save="no", status=36) # error, all windows in a feature files must have the same size | |
87 } | |
88 #error loading data | |
89 write("IWTomics message: An error has occurred reading the data. Please try again.", stderr()) | |
90 quit(save="no", status=30) | |
91 }) | |
92 | |
93 # smooth data | |
94 if(smoothing!='no'){ | |
95 tryCatch({ | |
96 if(smoothing=='locpoly'){ | |
97 dist_knots=10 | |
98 }else if(smoothing=='kernel'){ | |
99 degree=3 | |
100 dist_knots=10 | |
101 }else if(smoothing=='splines'){ | |
102 bandwidth=5 | |
103 } | |
104 if(alignment=='scale'){ | |
105 if(scale==0){ | |
106 regionsFeatures=smooth(regionsFeatures,type=smoothing,fill_gaps=fill_gaps, | |
107 bandwidth=bandwidth,degree=degree,dist_knots=dist_knots) | |
108 }else{ | |
109 regionsFeatures=smooth(regionsFeatures,type=smoothing,fill_gaps=fill_gaps, | |
110 bandwidth=bandwidth,degree=degree,dist_knots=dist_knots,scale_grid=scale) | |
111 } | |
112 }else{ | |
113 regionsFeatures=smooth(regionsFeatures,type=smoothing,fill_gaps=fill_gaps, | |
114 bandwidth=bandwidth,degree=degree,dist_knots=dist_knots) | |
115 } | |
116 }, error = function(err) { | |
117 write("IWTomics message: An error has occurred smoothing the data. Please try again.", stderr()) | |
118 quit(save="no", status=40) #error on smoothing | |
119 }) | |
120 } | |
121 | |
122 # plot data | |
123 pdf(outpdf,width=10,height=8) | |
124 if(plottype=='boxplot'){ | |
125 # fix repeated probs | |
126 probs=sort(unique(probs)) | |
127 }else{ | |
128 probs=c(0.25,0.5,0.75) | |
129 } | |
130 plot(regionsFeatures,type=plottype,probs=probs,average=average,size=size,ask=FALSE) | |
131 dev.off() | |
132 | |
133 # create output | |
134 #write.table(cbind(unlist(strsplit(args_values$regionsfilenames,'\\|')),idRegions(regionsFeatures),nameRegions(regionsFeatures)), | |
135 #file=outregions,quote=FALSE,sep='\t',row.names=FALSE,col.names=FALSE) | |
136 write.table(as.data.frame(t(idRegions(regionsFeatures))),file=outregions,quote=FALSE,sep='\t',row.names=FALSE,col.names=FALSE) | |
137 #write.table(cbind(unlist(strsplit(args_values$featuresfilenames,'\\|')),idFeatures(regionsFeatures),nameFeatures(regionsFeatures)), | |
138 #file=outfeatures,quote=FALSE,sep='\t',row.names=FALSE,col.names=FALSE) | |
139 write.table(as.data.frame(t(idFeatures(regionsFeatures))),file=outfeatures,quote=FALSE,sep='\t',row.names=FALSE,col.names=FALSE) | |
140 save(regionsFeatures,file=outrdata) | |
141 }else{ | |
142 write("IWTomics message: Missing IWTomics package. Please be sure to have it installed before using this tool.", stderr()) | |
143 quit(save="no", status=255) | |
144 } |