comparison SAINT_preprocessing_v6_mq_pep.py @ 9:9c3d2b1d0c1c draft

Uploaded
author bornea
date Tue, 10 Nov 2015 13:15:41 -0500
parents
children 729aac64ce43
comparison
equal deleted inserted replaced
8:30d53a378141 9:9c3d2b1d0c1c
1 #######################################################################################
2 # Python-code: SAINT pre-processing from maxquant "Samples Report" output
3 # Author: Brent Kuenzi
4 #######################################################################################
5 # This program reads in a raw maxquant "Samples Report" output and a user generated
6 # bait file and autoformats it into prey and interaction files for SAINTexpress
7 # analysis
8 #######################################################################################
9 import sys
10 import urllib2
11 import os
12 #######################################################################################
13 ## REQUIRED INPUT ##
14
15 # 1) infile: maxquant "Samples Report" output
16 # 2) baitfile: SAINT formatted bait file generated in Galaxy
17 # 3) prey: Y or N for generating a prey file (requires internet connection)
18 #######################################################################################
19 mq_file = sys.argv[1]
20 cmd = r"Rscript /home/bornea/galaxy_moffitt_dev/tools/Moffitt_Tools/bubblebeam/pre_process_protein_name_set.R " + str(mq_file)
21 os.system(cmd)
22
23 infile = "./tukeys_output.txt" #maxquant "Samples Report" output
24 prey = sys.argv[2] # Y or N
25 fasta_db = sys.argv[3]
26 if fasta_db == "None":
27 fasta_db = "/home/bornea/galaxy_moffitt_dev/tools/Moffitt_Tools/bubblebeam/SwissProt_HUMAN_2014_08.fasta"
28 make_bait= sys.argv[6]
29
30 def bait_create(baits, infile):
31 #Takes the Bait specified by the user and makes them into a Bait file and includes a check to make sure they are using valid baits.
32 baits = make_bait.split()
33 i = 0
34 bait_file_tmp = open("bait.txt", "wr")
35 order = []
36 bait_cache = []
37 while i < len(baits):
38 if baits[i+2] == "true":
39 T_C = "C"
40 else:
41 T_C = "T"
42 line1 = baits[i] + "\t" + baits[i+1] + "\t" + T_C + "\n"
43 q = open(infile,"r")
44 for line2 in q:
45 line2 = line2.replace("\"", "")
46 line2 = line2.replace(r"Intensity.", "") #R coerces "-" into "." this changes them back and remove Intensity from the Bait names.
47 line2 = line2.replace(r".", r"-")
48 temp = line2.split()
49 if "mapped_protein" in str(temp):
50 #If the bait is in the original file then write to cache it if not exit.
51 if baits[i] in temp:
52 number_bait = temp.index(str(baits[i]))
53 number_bait = number_bait - 9
54 bait_cache.append((number_bait, str(line1)))
55 else:
56 print "Error: bad bait " + str(baits[i])
57 sys.exit()
58 else:
59 pass
60 i = i + 3
61 #Writes cache to file.
62 bait_cache.sort()
63 for line in bait_cache:
64 bait_file_tmp.write(line[1])
65
66 bait_file_tmp.close()
67
68
69 baitfile = "bait.txt"
70
71 class ReturnValue1(object):
72 def __init__(self, sequence, gene):
73 self.seqlength = sequence
74 self.genename = gene
75 class ReturnValue2(object):
76 def __init__(self, getdata, getproteins, getheader):
77 self.data = getdata
78 self.proteins = getproteins
79 self.header = getheader
80
81 def main(maxquant_input, make_bait):
82 bait_create(make_bait, infile)
83 baitfile = "bait.txt"
84 #bait_check(baitfile, maxquant_input)
85 make_inter(maxquant_input)
86 if prey == 'true':
87 make_prey(maxquant_input)
88 no_error_inter(maxquant_input)
89 os.rename('prey.txt', sys.argv[5])
90 elif prey == 'false':
91 if os.path.isfile('error proteins.txt') == True:
92 no_error_inter(maxquant_input)
93 pass
94 elif prey != 'true' or 'false':
95 sys.exit("Invalid Prey Argument: Y or N")
96 os.rename('inter.txt', sys.argv[4])
97 os.rename("bait.txt", sys.argv[7])
98
99
100 def get_info(uniprot_accession_in): # get aa lengths and gene name
101 error = open('error proteins.txt', 'a+')
102 # while True:
103 # i = 0
104 # try:
105 # data = urllib2.urlopen("http://www.uniprot.org/uniprot/" + uniprot_accession_in + ".fasta")
106 # break
107 # except urllib2.HTTPError, err:
108 # i = i + 1
109 # if i == 50:
110 # sys.exit("More than 50 errors. Check your file or try again later.")
111 # if err.code == 404:
112 # error.write(uniprot_accession_in + '\t' + "Invalid URL. Check protein" + '\n')
113 # seqlength = 'NA'
114 # genename = 'NA'
115 # return ReturnValue1(seqlength, genename)
116 # elif err.code == 302:
117 # sys.exit("Request timed out. Check connection and try again.")
118 # else:
119 # sys.exit("Uniprot had some other error")
120 #
121 #
122 # if lines == []:
123 # error.write(uniprot_accession_in + '\t' + "Blank Fasta" + '\n')
124 # error.close
125 # if lines == []:
126 # error.write(uniprot_accession_in + '\t' + "Blank Fasta" + '\n')
127 # error.close
128 # seqlength = 'NA'
129 # genename = 'NA'
130 # return ReturnValue1(seqlength, genename)
131 # if lines != []:
132 # seqlength = 0
133 # header = lines[0]
134 # for line in lines[1:]:
135 # line = line.replace("\n","") # strip \n or else it gets counted in the length
136 # seqlength += len(line)
137 # if 'GN=' in header:
138 # lst = header.split('GN=')
139 # lst2 = lst[1].split(' ')
140 # genename = lst2[0]
141 # error.close
142 # return ReturnValue1(seqlength, genename)
143 # if 'GN=' not in header:
144 # genename = 'NA'
145 # error.close
146 # return ReturnValue1(seqlength, genename)
147
148 data = open(fasta_db,'r')
149 lines = data.readlines()
150 db_len = len(lines)
151 seqlength = 0
152 count = 0
153 for i in lines:
154 if ">sp" in i:
155 if uniprot_accession_in == i.split("|")[1]:
156 match = count+1
157 if 'GN=' in i:
158 lst = i.split('GN=')
159 lst2 = lst[1].split(' ')
160 genename = lst2[0]
161 if 'GN=' not in i:
162 genename = 'NA'
163 while ">sp" not in lines[match]:
164 if match <= db_len:
165 seqlength = seqlength + len(lines[match].strip())
166 match = match + 1
167 else:
168 break
169 return ReturnValue1(seqlength, genename)
170 count = count + 1
171
172
173 if seqlength == 0:
174 error.write(uniprot_accession_in + '\t' + "Uniprot not in Fasta" + '\n')
175 error.close
176 seqlength = 'NA'
177 genename = 'NA'
178 return ReturnValue1(seqlength, genename)
179
180
181 def readtab(infile):
182 with open(infile,'r') as x: # read in tab-delim text
183 output = []
184 for line in x:
185 line = line.strip()
186 temp = line.split('\t')
187 output.append(temp)
188 return output
189 def read_maxquant(maxquant_input): # Get data, proteins and header from maxquant output
190 dupes = readtab(maxquant_input)
191 header_start = 0
192 header = dupes[header_start]
193 for i in header:
194 i = i.replace(r"\"", "")
195 i = i.replace(r"Intensity.", r"")
196 i = i.replace(r".", r"-")
197 data = dupes[header_start+1:len(dupes)] #cut off blank line and END OF FILE
198 proteins = []
199 for protein in data:
200 proteins.append(protein[0])
201 return ReturnValue2(data, proteins, header)
202 def make_inter(maxquant_input):
203 bait = readtab(baitfile)
204 data = read_maxquant(maxquant_input).data
205 header = read_maxquant(maxquant_input).header
206 proteins = read_maxquant(maxquant_input).proteins
207 bait_index = []
208 for i in bait:
209 bait_index.append(header.index("mapped_protein") + 1) # Find just the baits defined in bait file
210 with open('inter.txt', 'w') as y:
211 a = 0; l=0
212 for bb in bait:
213 for lst in data:
214 y.write(header[bait_index[l]] + '\t' + bb[1] + '\t' + proteins[a] + '\t' + lst[bait_index[l]] + '\n')
215 a+=1
216 if a == len(proteins):
217 a = 0; l+=1
218 def make_prey(maxquant_input):
219 proteins = read_maxquant(maxquant_input).proteins
220 output_file = open("prey.txt",'w')
221 for a in proteins:
222 a = a.replace("\n","") # remove \n for input into function
223 a = a.replace("\r","") # ditto for \r
224 seq = get_info(a).seqlength
225 GN = get_info(a).genename
226 if seq != 'NA':
227 output_file.write(a+"\t"+str(seq)+ "\t" + str(GN) + "\n")
228 output_file.close()
229 def no_error_inter(maxquant_input): # remake inter file without protein errors from Uniprot
230 err = readtab("error proteins.txt")
231 bait = readtab(baitfile)
232 data = read_maxquant(maxquant_input).data
233 header = read_maxquant(maxquant_input).header
234 header = [i.replace(r"\"", "") for i in header]
235 header = [i.replace(r"Intensity.", r"") for i in header]
236 header = [i.replace(r".", r"-") for i in header]
237 print header
238 bait_index = []
239 for i in bait:
240 bait_index.append(header.index(i[0]))
241 proteins = read_maxquant(maxquant_input).proteins
242 errors = []
243 for e in err:
244 errors.append(e[0])
245 with open('inter.txt', 'w') as y:
246 l = 0; a = 0
247 for bb in bait:
248 for lst in data:
249 if proteins[a] not in errors:
250 y.write(header[bait_index[l]] + '\t' + bb[1] + '\t' + proteins[a] + '\t' + lst[bait_index[l]] + '\n')
251 a+=1
252 if a == len(proteins):
253 l += 1; a = 0
254 def bait_check(bait, maxquant_input): # check that bait names share header titles
255 bait_in = readtab(bait)
256 header = read_maxquant(maxquant_input).header
257 for i in bait_in:
258 if i[0] not in header:
259 sys.exit("Bait must share header titles with maxquant output")
260
261 if __name__ == '__main__':
262 main(infile, make_bait)