| 
26
 | 
     1 #######################################################################################
 | 
| 
 | 
     2 # Python-code: SAINT pre-processing from Scaffold "Samples Report" output
 | 
| 
 | 
     3 # Author: Brent Kuenzi
 | 
| 
 | 
     4 #######################################################################################
 | 
| 
 | 
     5 # This program reads in a raw Scaffold "Samples Report" output and a user generated
 | 
| 
 | 
     6 # bait file and autoformats it into prey and interaction files for SAINTexpress
 | 
| 
 | 
     7 # analysis
 | 
| 
 | 
     8 #######################################################################################
 | 
| 
 | 
     9 # Copyright (C)  Brent Kuenzi.
 | 
| 
 | 
    10 # Permission is granted to copy, distribute and/or modify this document
 | 
| 
 | 
    11 # under the terms of the GNU Free Documentation License, Version 1.3
 | 
| 
 | 
    12 # or any later version published by the Free Software Foundation;
 | 
| 
 | 
    13 # with no Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts.
 | 
| 
 | 
    14 # A copy of the license is included in the section entitled "GNU
 | 
| 
 | 
    15 # Free Documentation License".
 | 
| 
 | 
    16 #######################################################################################
 | 
| 
 | 
    17 ## REQUIRED INPUT ##
 | 
| 
 | 
    18 
 | 
| 
 | 
    19 # 1) infile: Scaffold "Samples Report" output
 | 
| 
 | 
    20 # 2) baitfile: SAINT formatted bait file generated in Galaxy
 | 
| 
 | 
    21 # 3) fasta_db: fasta database for use (defaults to SwissProt_HUMAN_2014_08.fasta)
 | 
| 
 | 
    22 # 4) prey: Y or N for generating a prey file
 | 
| 
 | 
    23 # 5) make_bait: String of bait names, assignment, and test or control boolean
 | 
| 
 | 
    24 #######################################################################################
 | 
| 
 | 
    25 
 | 
| 
 | 
    26 import sys
 | 
| 
 | 
    27 import os.path
 | 
| 
 | 
    28 
 | 
| 
 | 
    29 
 | 
| 
 | 
    30 infile = sys.argv[1] 
 | 
| 
 | 
    31 #Scaffold "Samples Report" output.
 | 
| 
 | 
    32 prey = sys.argv[2] 
 | 
| 
 | 
    33 # Y or N boolean from Galaxy.
 | 
| 
 | 
    34 fasta_db = sys.argv[3]
 | 
| 
 | 
    35 tool_path = sys.argv[8]
 | 
| 
 | 
    36 if fasta_db == "None":
 | 
| 
 | 
    37     fasta_db = str(tool_path)  + "/SwissProt_HUMAN_2014_08.fasta"
 | 
| 
 | 
    38 make_bait = sys.argv[6]
 | 
| 
 | 
    39 bait_bool = sys.argv[9]
 | 
| 
 | 
    40 
 | 
| 
 | 
    41 
 | 
| 
 | 
    42 def bait_create(baits, infile):
 | 
| 
 | 
    43     # Verifies the Baits are valid in the Scaffold file and writes the Bait.txt.
 | 
| 
 | 
    44     baits = make_bait.split()
 | 
| 
 | 
    45     i = 0
 | 
| 
 | 
    46     bait_file_tmp = open("bait.txt", "w")
 | 
| 
 | 
    47     order = []
 | 
| 
 | 
    48     bait_cache = []
 | 
| 
 | 
    49     while i < len(baits):
 | 
| 
 | 
    50         if baits[i+2] == "true":
 | 
| 
 | 
    51             T_C = "C"
 | 
| 
 | 
    52         else:
 | 
| 
 | 
    53             T_C = "T"
 | 
| 
 | 
    54         bait_line = baits[i] + "\t" + baits[i+1] + "\t" + T_C + "\n"
 | 
| 
 | 
    55         read_infile = open(infile, "r")
 | 
| 
 | 
    56         for input_line in read_infile:
 | 
| 
 | 
    57             input_line = input_line.strip()
 | 
| 
 | 
    58             temp = input_line.split('\t')
 | 
| 
 | 
    59             if "Quantitative Variance" in str(temp):
 | 
| 
 | 
    60                 if baits[i] in temp:                    
 | 
| 
 | 
    61                     number_bait = temp.index(str(baits[i]))
 | 
| 
 | 
    62                     number_bait = number_bait - 9
 | 
| 
 | 
    63                     bait_cache.append((number_bait, str(bait_line)))
 | 
| 
 | 
    64                     # Locates the Bait names in the column names and then sets the Baits in the 
 | 
| 
 | 
    65                     # correct order in the cache thus the - 9 because the baits start at the 9th
 | 
| 
 | 
    66                     # column.
 | 
| 
 | 
    67                 else:
 | 
| 
 | 
    68                     print "Error: bad bait " + str(baits[i])
 | 
| 
 | 
    69                     sys.exit()
 | 
| 
 | 
    70             else:
 | 
| 
 | 
    71                 pass
 | 
| 
 | 
    72         i = i + 3
 | 
| 
 | 
    73 
 | 
| 
 | 
    74     bait_cache.sort()
 | 
| 
 | 
    75     for cache_line in bait_cache:
 | 
| 
 | 
    76         bait_file_tmp.write(cache_line[1])
 | 
| 
 | 
    77 
 | 
| 
 | 
    78     bait_file_tmp.close()
 | 
| 
 | 
    79 
 | 
| 
 | 
    80 if bait_bool == 'false':
 | 
| 
 | 
    81     bait_create(make_bait, infile)
 | 
| 
 | 
    82     baitfile = "bait.txt"
 | 
| 
 | 
    83 else:
 | 
| 
 | 
    84     bait_temp_file = open(sys.argv[10], 'r')
 | 
| 
 | 
    85     bait_cache = bait_temp_file.readlines()
 | 
| 
 | 
    86     print bait_cache
 | 
| 
 | 
    87     bait_file_tmp = open("bait.txt", "wr")
 | 
| 
 | 
    88     for cache_line in bait_cache:
 | 
| 
 | 
    89         bait_file_tmp.write(cache_line)
 | 
| 
 | 
    90     bait_file_tmp.close()
 | 
| 
 | 
    91     baitfile = "bait.txt"
 | 
| 
 | 
    92 
 | 
| 
 | 
    93 
 | 
| 
 | 
    94 class ReturnValue1(object):
 | 
| 
 | 
    95     def __init__(self, sequence, gene):
 | 
| 
 | 
    96         self.seqlength = sequence
 | 
| 
 | 
    97         self.genename = gene
 | 
| 
 | 
    98 
 | 
| 
 | 
    99 
 | 
| 
 | 
   100 class ReturnValue2(object):
 | 
| 
 | 
   101     def __init__(self, getdata, getproteins, getheader):
 | 
| 
 | 
   102         self.data = getdata
 | 
| 
 | 
   103         self.proteins = getproteins
 | 
| 
 | 
   104         self.header = getheader
 | 
| 
 | 
   105 
 | 
| 
 | 
   106 
 | 
| 
 | 
   107 def main(Scaffold_input, baits):
 | 
| 
 | 
   108     bait_check(baitfile, Scaffold_input)
 | 
| 
 | 
   109     make_inter(Scaffold_input)
 | 
| 
 | 
   110     if prey == 'true':
 | 
| 
 | 
   111         make_prey(Scaffold_input)
 | 
| 
 | 
   112         no_error_inter(Scaffold_input)
 | 
| 
 | 
   113         os.rename('prey.txt', sys.argv[5])
 | 
| 
 | 
   114     elif prey == 'false':
 | 
| 
 | 
   115         if os.path.isfile('error proteins.txt') == True:
 | 
| 
 | 
   116             no_error_inter(Scaffold_input)
 | 
| 
 | 
   117         pass
 | 
| 
 | 
   118     elif prey != 'true' or 'false':
 | 
| 
 | 
   119         sys.exit("Invalid Prey Argument: Y or N")
 | 
| 
 | 
   120 
 | 
| 
 | 
   121 
 | 
| 
 | 
   122 def get_info(uniprot_accession_in): 
 | 
| 
 | 
   123     # Get aminoacid lengths and gene name.
 | 
| 
 | 
   124     error = open('error proteins.txt', 'a+')
 | 
| 
 | 
   125     data = open(fasta_db, 'r')
 | 
| 
 | 
   126     data_lines = data.readlines()
 | 
| 
 | 
   127     db_len = len(data_lines)
 | 
| 
 | 
   128     seqlength = 0
 | 
| 
 | 
   129     count = 0
 | 
| 
 | 
   130     for data_line in data_lines:
 | 
| 
 | 
   131         if ">sp" in data_line:
 | 
| 
 | 
   132             namer = data_line.split("|")[2]
 | 
| 
 | 
   133         if uniprot_accession_in == data_line.split("|")[1]:
 | 
| 
 | 
   134             match = count + 1
 | 
| 
 | 
   135             if 'GN=' in data_line:
 | 
| 
 | 
   136                 lst = data_linedata_line.split('GN=')
 | 
| 
 | 
   137                 lst2 = lst[1].split(' ')
 | 
| 
 | 
   138                 genename = lst2[0]
 | 
| 
 | 
   139             if 'GN=' not in data_line:
 | 
| 
 | 
   140                 genename = 'NA'
 | 
| 
 | 
   141             while ">sp" not in data_lines[match]:
 | 
| 
 | 
   142                 if match <= db_len:
 | 
| 
 | 
   143                     seqlength = seqlength + len(data_lines[match].strip())
 | 
| 
 | 
   144                     match = match + 1
 | 
| 
 | 
   145                 else:
 | 
| 
 | 
   146                     break
 | 
| 
 | 
   147             return ReturnValue1(seqlength, genename)
 | 
| 
 | 
   148         elif uniprot_accession_in == namer.split(" ")[0]:
 | 
| 
 | 
   149             match = count + 1
 | 
| 
 | 
   150             # Ensures consistent spacing throughout.
 | 
| 
 | 
   151             if 'GN=' in data_line:
 | 
| 
 | 
   152                 lst = data_line.split('GN=')
 | 
| 
 | 
   153                 lst2 = lst[1].split(' ')
 | 
| 
 | 
   154                 genename = lst2[0]
 | 
| 
 | 
   155             if 'GN=' not in data_line:
 | 
| 
 | 
   156                 genename = 'NA'
 | 
| 
 | 
   157             while ">sp" not in lines[match]:
 | 
| 
 | 
   158                 if match <= db_len:
 | 
| 
 | 
   159                     seqlength = seqlength + len(lines[match].strip())
 | 
| 
 | 
   160                     match = match + 1
 | 
| 
 | 
   161                 else:
 | 
| 
 | 
   162                     break
 | 
| 
 | 
   163             return ReturnValue1(seqlength, genename)
 | 
| 
 | 
   164         count = count + 1
 | 
| 
 | 
   165     if seqlength == 0:
 | 
| 
 | 
   166         error.write(uniprot_accession_in + '\t' + "Uniprot not in Fasta" + '\n')
 | 
| 
 | 
   167         error.close
 | 
| 
 | 
   168         seqlength = 'NA'
 | 
| 
 | 
   169         genename = 'NA'
 | 
| 
 | 
   170         return ReturnValue1(seqlength, genename)
 | 
| 
 | 
   171 
 | 
| 
 | 
   172 
 | 
| 
 | 
   173 def readtab(infile):
 | 
| 
 | 
   174     with open(infile, 'r') as input_file: 
 | 
| 
 | 
   175     # read in tab-delim text
 | 
| 
 | 
   176         output = []
 | 
| 
 | 
   177         for input_line in input_file:
 | 
| 
 | 
   178             input_line = input_line.strip()
 | 
| 
 | 
   179             temp = input_line.split('\t')
 | 
| 
 | 
   180             output.append(temp)
 | 
| 
 | 
   181     return output
 | 
| 
 | 
   182 
 | 
| 
 | 
   183 
 | 
| 
 | 
   184 def read_Scaffold(Scaffold_input): 
 | 
| 
 | 
   185     # Get data, proteins and header from Scaffold output
 | 
| 
 | 
   186     dupes = readtab(Scaffold_input)
 | 
| 
 | 
   187     cnt = 0
 | 
| 
 | 
   188     for Scaffold_line in dupes:
 | 
| 
 | 
   189         cnt += 1
 | 
| 
 | 
   190         if Scaffold_line[0] == '#': 
 | 
| 
 | 
   191         # Finds the start of second header.
 | 
| 
 | 
   192             header_start = cnt-1
 | 
| 
 | 
   193     header = dupes[header_start]
 | 
| 
 | 
   194     prot_start = header.index("Accession Number")
 | 
| 
 | 
   195     data = dupes[header_start+1:len(dupes)-2] 
 | 
| 
 | 
   196     # Cut off blank line and END OF FILE.
 | 
| 
 | 
   197     proteins = []
 | 
| 
 | 
   198     for Scaffold_line in data:
 | 
| 
 | 
   199         Scaffold_line[4] = Scaffold_line[4].split()[0]
 | 
| 
 | 
   200         # Removes the (+##) that sometimes is attached.
 | 
| 
 | 
   201     for protein in data:
 | 
| 
 | 
   202         proteins.append(protein[prot_start])
 | 
| 
 | 
   203     return ReturnValue2(data, proteins, header)
 | 
| 
 | 
   204 
 | 
| 
 | 
   205 
 | 
| 
 | 
   206 def make_inter(Scaffold_input):
 | 
| 
 | 
   207     bait = readtab(baitfile)
 | 
| 
 | 
   208     data = read_Scaffold(Scaffold_input).data
 | 
| 
 | 
   209     header = read_Scaffold(Scaffold_input).header
 | 
| 
 | 
   210     proteins = read_Scaffold(Scaffold_input).proteins
 | 
| 
 | 
   211     bait_index = []
 | 
| 
 | 
   212     for bait_line in bait:
 | 
| 
 | 
   213         bait_index.append(header.index(bait_line[0]))
 | 
| 
 | 
   214         # Find just the baits defined in bait file.
 | 
| 
 | 
   215     with open('inter.txt', 'w') as inter_file:
 | 
| 
 | 
   216         a = 0; l = 0
 | 
| 
 | 
   217         for bb in bait:
 | 
| 
 | 
   218             for lst in data:
 | 
| 
 | 
   219                 inter_file.write(header[bait_index[l]] + '\t' + bb[1] + '\t' + proteins[a] + '\t'
 | 
| 
 | 
   220                         + lst[bait_index[l]] + '\n')
 | 
| 
 | 
   221                 a += 1
 | 
| 
 | 
   222                 if a == len(proteins):
 | 
| 
 | 
   223                     a = 0; l += 1
 | 
| 
 | 
   224 
 | 
| 
 | 
   225 
 | 
| 
 | 
   226 def make_prey(Scaffold_input):
 | 
| 
 | 
   227     proteins = read_Scaffold(Scaffold_input).proteins
 | 
| 
 | 
   228     output_file = open("prey.txt", 'w')
 | 
| 
 | 
   229     for protein in proteins:
 | 
| 
 | 
   230         protein = protein.replace("\n", "")
 | 
| 
 | 
   231         # Remove \n for input into function.
 | 
| 
 | 
   232         protein = protein.replace("\r", "")
 | 
| 
 | 
   233         # Ditto for \r.
 | 
| 
 | 
   234         seq = get_info(protein).seqlength
 | 
| 
 | 
   235         GN = get_info(protein).genename
 | 
| 
 | 
   236         if seq != 'NA':
 | 
| 
 | 
   237             output_file.write(protein + "\t" + str(seq) + "\t" + str(GN) + "\n")
 | 
| 
 | 
   238     output_file.close()
 | 
| 
 | 
   239 
 | 
| 
 | 
   240 
 | 
| 
 | 
   241 def no_error_inter(Scaffold_input):
 | 
| 
 | 
   242     # Remake inter file without protein errors from Uniprot.
 | 
| 
 | 
   243     err = readtab("error proteins.txt")
 | 
| 
 | 
   244     bait = readtab(baitfile)
 | 
| 
 | 
   245     data = read_Scaffold(Scaffold_input).data
 | 
| 
 | 
   246     header = read_Scaffold(Scaffold_input).header
 | 
| 
 | 
   247     bait_index = []
 | 
| 
 | 
   248     for bait_line in bait:
 | 
| 
 | 
   249         bait_index.append(header.index(bait_line[0]))
 | 
| 
 | 
   250     proteins = read_Scaffold(Scaffold_input).proteins
 | 
| 
 | 
   251     errors = []
 | 
| 
 | 
   252     for e in err:
 | 
| 
 | 
   253         errors.append(e[0])
 | 
| 
 | 
   254     with open('inter.txt', 'w') as y:
 | 
| 
 | 
   255         l = 0; a = 0
 | 
| 
 | 
   256         for bb in bait:
 | 
| 
 | 
   257             for lst in data:
 | 
| 
 | 
   258                 if proteins[a] not in errors:
 | 
| 
 | 
   259                     y.write(header[bait_index[l]] + '\t' + bb[1] + '\t' + proteins[a] + '\t'
 | 
| 
 | 
   260                             + lst[bait_index[l]] + '\n')
 | 
| 
 | 
   261                 a += 1
 | 
| 
 | 
   262                 if a == len(proteins):
 | 
| 
 | 
   263                     l += 1; a = 0
 | 
| 
 | 
   264 
 | 
| 
 | 
   265 
 | 
| 
 | 
   266 def bait_check(bait, Scaffold_input): 
 | 
| 
 | 
   267     # Check that bait names share Scaffold header titles.
 | 
| 
 | 
   268     bait_in = readtab(bait)
 | 
| 
 | 
   269     header = read_Scaffold(Scaffold_input).header
 | 
| 
 | 
   270     for i in bait_in:
 | 
| 
 | 
   271         if i[0] not in header:
 | 
| 
 | 
   272             sys.exit("Bait must share header titles with Scaffold output")
 | 
| 
 | 
   273 
 | 
| 
 | 
   274 if __name__ == '__main__':
 | 
| 
 | 
   275     main(infile, baitfile)
 | 
| 
 | 
   276 
 | 
| 
 | 
   277 os.rename("inter.txt", sys.argv[4])
 | 
| 
 | 
   278 os.rename("bait.txt", sys.argv[7])
 |