diff combine_output.py @ 0:9d5f4f5f764b

Initial commit to toolshed
author pieter.lukasse@wur.nl
date Thu, 16 Jan 2014 13:10:00 +0100
parents
children 19d8fd10248e
line wrap: on
line diff
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/combine_output.py	Thu Jan 16 13:10:00 2014 +0100
@@ -0,0 +1,230 @@
+#!/usr/bin/env python
+# encoding: utf-8
+'''
+Module to combine output from two GCMS Galaxy tools (RankFilter and CasLookup)
+'''
+
+import csv
+import re
+import sys
+import math
+import pprint
+
+__author__ = "Marcel Kempenaar"
+__contact__ = "brs@nbic.nl"
+__copyright__ = "Copyright, 2012, Netherlands Bioinformatics Centre"
+__license__ = "MIT"
+
+def _process_data(in_csv):
+    '''
+    Generic method to parse a tab-separated file returning a dictionary with named columns
+    @param in_csv: input filename to be parsed
+    '''
+    data = list(csv.reader(open(in_csv, 'rU'), delimiter='\t'))
+    header = data.pop(0)
+    # Create dictionary with column name as key
+    output = {}
+    for index in xrange(len(header)):
+        output[header[index]] = [row[index] for row in data]
+    return output
+
+
+def _merge_data(rankfilter, caslookup):
+    '''
+    Merges data from both input dictionaries based on the Centrotype field. This method will
+    build up a new list containing the merged hits as the items. 
+    @param rankfilter: dictionary holding RankFilter output in the form of N lists (one list per attribute name)
+    @param caslookup: dictionary holding CasLookup output in the form of N lists (one list per attribute name)
+    '''
+    # TODO: test for correct input files -> rankfilter and caslookup internal lists should have the same lenghts:
+    if (len(rankfilter['ID']) != len(caslookup['Centrotype'])):
+        raise Exception('rankfilter and caslookup files should have the same nr of rows/records ')
+    
+    merged = []
+    processed = {}
+    for compound_id_idx in xrange(len(rankfilter['ID'])):
+        compound_id = rankfilter['ID'][compound_id_idx]
+        if not compound_id in processed :
+            # keep track of processed items to not repeat them
+            processed[compound_id] = compound_id
+            # get centrotype nr
+            centrotype = compound_id.split('-')[0]
+            # Get the indices for current compound ID in both data-structures for proper matching
+            rindex = [index for index, value in enumerate(rankfilter['ID']) if value == compound_id]
+            cindex = [index for index, value in enumerate(caslookup['Centrotype']) if value == centrotype]
+            
+            merged_hits = []
+            # Combine hits
+            for hit in xrange(len(rindex)):
+                # Create records of hits to be merged ("keys" are the attribute names, so what the lines below do 
+                # is create a new "dict" item with same "keys"/attributes, with each attribute filled with its
+                # corresponding value in the rankfilter or caslookup tables; i.e. 
+                # rankfilter[key] => returns the list/array with size = nrrows, with the values for the attribute
+                #                    represented by "key". rindex[hit] => points to the row nr=hit (hit is a rownr/index)
+                rf_record = dict(zip(rankfilter.keys(), [rankfilter[key][rindex[hit]] for key in rankfilter.keys()]))
+                cl_record = dict(zip(caslookup.keys(), [caslookup[key][cindex[hit]] for key in caslookup.keys()]))
+                
+                merged_hit = _add_hit(rf_record, cl_record)
+                merged_hits.append(merged_hit)
+                
+            merged.append(merged_hits)
+
+    return merged, len(rindex)
+
+
+def _add_hit(rankfilter, caslookup):
+    '''
+    Combines single records from both the RankFilter- and CasLookup-tools
+    @param rankfilter: record (dictionary) of one compound in the RankFilter output
+    @param caslookup: matching record (dictionary) of one compound in the CasLookup output
+    '''
+    # The ID in the RankFilter output contains the following 5 fields:
+    rf_id = rankfilter['ID'].split('-')
+    try:
+        name, formula = _remove_formula(rankfilter['Name'])
+        hit = [rf_id[0], # Centrotype
+               rf_id[1], # cent.Factor
+               rf_id[2], # scan nr
+               rf_id[3], # R.T. (umin)
+               rf_id[4], # nr. Peaks
+               # Appending other fields
+               rankfilter['R.T.'],
+               name,
+               caslookup['FORMULA'] if not formula else formula,
+               rankfilter['Library'].strip(),
+               rankfilter['CAS'].strip(),
+               rankfilter['Forward'],
+               rankfilter['Reverse'],
+               ((float(rankfilter['Forward']) + float(rankfilter['Reverse'])) / 2),
+               rankfilter['RIexp'],
+               caslookup['RI'],
+               rankfilter['RIsvr'],
+               # Calculate absolute differences
+               math.fabs(float(rankfilter['RIexp']) - float(rankfilter['RIsvr'])),
+               math.fabs(float(caslookup['RI']) - float(rankfilter['RIexp'])),
+               caslookup['Regression.Column.Name'],
+               caslookup['min'],
+               caslookup['max'],
+               caslookup['nr.duplicates'],
+               caslookup['Column.phase.type'],
+               caslookup['Column.name'],
+               rankfilter['Rank'],
+               rankfilter['%rel.err'],
+               rankfilter['Synonyms']]
+    except KeyError as error:
+        print "Problem reading in data from input file(s):\n",
+        print "Respective CasLookup entry: \n", pprint.pprint(caslookup), "\n"
+        print "Respective RankFilter entry: \n", pprint.pprint(rankfilter), "\n"
+        raise error
+
+    return hit
+
+
+def _remove_formula(name):
+    '''
+    The RankFilter Name field often contains the Formula as well, this function removes it from the Name
+    @param name: complete name of the compound from the RankFilter output
+    '''
+    name = name.split()
+    poss_formula = name[-1]
+    match = re.match("^(([A-Z][a-z]{0,2})(\d*))+$", poss_formula)
+    if match:
+        return ' '.join(name[:-1]), poss_formula
+    else:
+        return ' '.join(name), False
+
+
+def _get_default_caslookup():
+    '''
+    The Cas Lookup tool might not have found all compounds in the library searched,
+    this default dict will be used to combine with the Rank Filter output
+    '''
+    return {'FORMULA': 'N/A',
+            'RI': '0.0',
+            'Regression.Column.Name': 'None',
+            'min': '0.0',
+            'max': '0.0',
+            'nr.duplicates': '0',
+            'Column.phase.type': 'N/A',
+            'Column.name': 'N/A'}
+
+
+def _save_data(data, nhits, out_csv_single, out_csv_multi):
+    '''
+    Writes tab-separated data to file
+    @param data: dictionary containing merged dataset
+    @param out_csv: output csv file
+    '''
+    header = ['Centrotype',
+              'cent.Factor',
+              'scan nr.',
+              'R.T. (umin)',
+              'nr. Peaks',
+              'R.T.',
+              'Name',
+              'FORMULA',
+              'Library',
+              'CAS',
+              'Forward',
+              'Reverse',
+              'Avg. (Forward, Reverse)',
+              'RIexp',
+              'RI',
+              'RIsvr',
+              'RIexp - RIsvr',
+              'RI - RIexp',
+              'Regression.Column.Name',
+              'min',
+              'max',
+              'nr.duplicates',
+              'Column.phase.type',
+              'Column.name',
+              'Rank',
+              '%rel.err',
+              'Synonyms']
+
+    # Open output file for writing
+    outfile_single_handle = open(out_csv_single, 'wb')
+    outfile_multi_handle = open(out_csv_multi, 'wb')
+    output_single_handle = csv.writer(outfile_single_handle, delimiter="\t")
+    output_multi_handle = csv.writer(outfile_multi_handle, delimiter="\t")
+
+    # Write headers
+    output_single_handle.writerow(header)
+    output_multi_handle.writerow(header * nhits)
+    # Combine all hits for each centrotype into one line
+    line = []
+    for centrotype_idx in xrange(len(data)):
+        for hit in data[centrotype_idx]:
+            line.extend(hit)
+        output_multi_handle.writerow(line)
+        line = []
+
+    # Write one line for each centrotype
+    for centrotype_idx in xrange(len(data)):
+        for hit in data[centrotype_idx]:
+            output_single_handle.writerow(hit)
+
+
+def main():
+    '''
+    Combine Output main function
+    It will merge the result files from "RankFilter"  and "Lookup RI for CAS numbers" 
+    NB: the caslookup_result_file will typically have fewer lines than
+    rankfilter_result_file, so the merge has to consider this as well. The final file
+    should have the same nr of lines as rankfilter_result_file.
+    '''
+    rankfilter_result_file = sys.argv[1]
+    caslookup_result_file = sys.argv[2]
+    output_single_csv = sys.argv[3]
+    output_multi_csv = sys.argv[4]
+
+    # Read RankFilter and CasLookup output files
+    rankfilter = _process_data(rankfilter_result_file)
+    caslookup = _process_data(caslookup_result_file)
+    merged, nhits = _merge_data(rankfilter, caslookup)
+    _save_data(merged, nhits, output_single_csv, output_multi_csv)
+
+
+if __name__ == '__main__':
+    main()