############################################################################## # Prerequisites for all functions # ############################################################################## import cando as cnd from compbench import * import argparse, os, random # If using command-line interface, set RUN_THROUGH = CLI # If running commands written in file, set RUN_THROUGH = FILE CLI = 0 FILE = 1 RUN_THROUGH = FILE def generate_cando_obj(matrix, cmpds, c2i, cmpd_set, ncpus=1, val='name'): ''' matrix - str, name of protein interaction signature matrix cmpds - str, name of drug file c2i - str, name of drug-indication mapping file cmpd_set - set (or list), set of compounds to be included ncpus - int, number of CPUs available for use by CANDO object Returns the CANDO object, an indication object, and cmpd_object dict Indication object can be used to rapidly create/test indications ''' cando_obj = cnd.CANDO(cmpds, c2i, matrix=matrix, compute_distance=True, dist_metric='cosine', ncpus=ncpus) test_indic = cnd.Indication('test_indic', 'Temp Test Indic') cando_obj.indications.append(test_indic) cando_obj.indication_ids.append('test_indic') print(len(cando_obj.compounds)) cmpd_to_obj = {} for cmpd in cmpd_set: if type(cmpd) == str and cmpd.isdigit(): obj = cando_obj.get_compound(int(cmpd), quiet=True) else: obj = cando_obj.get_compound(cmpd, quiet=True) cmpd_to_obj[cmpd] = obj return cando_obj, test_indic, cmpd_to_obj ############################################################################## # CANDO primary pipeline variants # ############################################################################## def cando_all_similar(indicated, non_indicated, cando_obj, test_indic, cmpd_to_obj, filename): ''' CANDO calculates ranks with similarity list cutoff (n) = -1 (compounds sorted by overall average rank, not consensus score) indicated - set of str, cmpds with some commonality ranked output is based on similarity to the indicated group non_indicated - set of str, cmpds to be ranked; mutually excl w/indicated may include indicated cmpds not included in indicated set for benchmarking cando_obj - CANDO object, instantiated with correct matrix/drug file indication mapping file does not matter b/c will use "indicated" set test_indic - Indication object, part of cando_obj will be altered to include indicated drugs & then assessed cmpd_to_obj - dictionary, str:Compound object used to populate test_indic with Compound objects based on indicated set filename - str, name of temporary output file (to avoid filename clashing) returns ranks - list of str, compounds ordered based on similarity to indicated compounds by CANDO ''' assert len(indicated.intersection(non_indicated)) == 0,\ 'Overlap found between indicated and non-indicated groups.' test_indic.compounds = [cmpd_to_obj[x] for x in list(indicated)] temp_file = 'temp_rank_%s_alt.tsv' % filename cando_obj.canpredict_compounds('test_indic', topX=-1, n=-1, consensus=False, save=temp_file) with open(temp_file, 'r') as f: lines = f.read().strip().split('\n')[1:] ranks = [] for line in lines: cells = line.split('\t') if type(list(indicated)[0]) == str and list(indicated)[0].isdigit(): cmpd = cells[4] else: cmpd = cells[6] if cmpd in non_indicated: ranks.append(cmpd) #os.remove(temp_file) # Comment this out to view the file generated return ranks def cando_multiple_lists(indicated, non_indicated, cando_obj, test_indic, cmpd_to_obj, filename): ''' CANDO calculates ranks with similarity list cutoff (n) = {10, 20, 30...} sorted by consensus score & average rank above similarity list cutoff compounds retrieved at later/greater similarity list cutoffs ranked after those retrieved at earlier/smaller similarity list cutoffs indicated - set of str, cmpds with some commonality ranked output is based on similarity to the indicated group non_indicated - set of str, cmpds to be ranked; mutually excl w/indicated may include indicated cmpds not included in indicated set for benchmarking cando_obj - CANDO object, instantiated with correct matrix/drug file indication mapping file does not matter b/c will use "indicated" set test_indic - Indication object, part of cando_obj will be altered to include indicated drugs & then assessed cmpd_to_obj - dictionary, str:Compound object used to populate test_indic with Compound objects based on indicated set filename - str, name of temporary output file (to avoid filename clashing) returns ranks - list of str, compounds ordered based on similarity to indicated compounds by CANDO ''' assert len(indicated.intersection(non_indicated)) == 0,\ 'Overlap found between indicated and non-indicated groups.' test_indic.compounds = [cmpd_to_obj[x] for x in list(indicated)] temp_file = 'temp_rank_%s.tsv' % filename n = 10 ranks = [] ranked = set() #print('Benchmarking %d cmpd indication...' % len(indicated)) try: while len(non_indicated - ranked) > 0: cando_obj.canpredict_compounds('test_indic', topX=-1, n=n, consensus=False, save=temp_file) with open(temp_file, 'r') as f: lines = f.read().strip().split('\n')[1:] for line in lines: cells = line.split('\t') if type(list(indicated)[0]) == str and list(indicated)[0].isdigit(): cmpd = cells[4] else: cmpd = cells[6] if cmpd in non_indicated and cmpd not in ranked: ranks.append(cmpd) ranked.add(cmpd) n += 10 except IndexError: # If something is not captured when iterating in units of 10 # ie only captured in last 9 n values unranked = list(non_indicated - ranked) random.shuffle(unranked) random.shuffle(unranked) random.shuffle(unranked) ranks += unranked os.remove(temp_file) # Comment this out to view the file generated return ranks def cando_ten_similar(indicated, non_indicated, cando_obj, test_indic, cmpd_to_obj, filename): ''' CANDO calculates ranks with similarity list cutoff (n) = 10 sorted by consensus score & average rank above similarity list cutoff compounds not retrieved w/n=10 ordered based on overall average rank indicated - set of str, cmpds with some commonality ranked output is based on similarity to the indicated group non_indicated - set of str, cmpds to be ranked; mutually excl w/indicated may include indicated cmpds not included in indicated set for benchmarking cando_obj - CANDO object, instantiated with correct matrix/drug file indication mapping file does not matter b/c will use "indicated" set test_indic - Indication object, part of cando_obj will be altered to include indicated drugs & then assessed cmpd_to_obj - dictionary, str:Compound object used to populate test_indic with Compound objects based on indicated set filename - str, name of temporary output file (to avoid filename clashing) returns ranks - list of str, compounds ordered based on similarity to indicated compounds by CANDO ''' assert len(indicated.intersection(non_indicated)) == 0,\ 'Overlap found between indicated and non-indicated groups.' test_indic.compounds = [cmpd_to_obj[x] for x in list(indicated)] #print('x', end='') # used to track progress n = 10 sorted_x = cando_obj.canpredict_compounds('test_indic', topX=-1, n=n, consensus=False, return_all=True) if list(indicated)[0].isdigit(): ranks = [str(cando_obj.get_compound(x[0]).id_) for x in sorted_x] else: ranks = [cando_obj.get_compound(x[0]).name for x in sorted_x] ranks = [x for x in ranks if x in non_indicated] return ranks ############################################################################## # Test/benchmarking code # ############################################################################## fileset = 'CTD' #FUNC = cando_all_similar FUNC = cando_ten_similar #FUNC = cando_multiple_lists out_name = 'cando_ten_similar_%s' % fileset if fileset.lower() == 'ttd': ctd_file = 'cando_data/drugbank2ctd-v2.5.tsv' # Req for CANDO obj, not used drug_file = 'cando_data/drugbank-v2.5_alt.tsv' matrix_file = 'cando_data/unlabelled_v2.5_all_CxP_matrix_alt.tsv' drug_indic_map = 'data/TTD_approved_drugs_cndonly.tsv' cross_val_file = 'data/name_10fold_cross_seed0_TTD.tsv' key,val='iname','dname' elif fileset.lower() == 'ctd': ctd_file = 'cando_data/drugbank2ctd-v2.5.tsv' drug_file = 'cando_data/drugbank-v2.5_ctdalt.tsv' # no unapproved/unindicated drugs matrix_file = 'cando_data/unlabelled_v2.5_all_CxP_matrix_ctdalt.tsv' drug_indic_map = 'data/ctd_approved_drugs.tsv' #cross_val_file = 'data/id_10fold_cross_seed0_ctd_cnd.tsv' cross_val_file = 'data/name_10fold_cross_seed0_ctd_cnd.tsv' #key,val='iid','did' key,val='iname','dname' else: raise ValueError('No fileset ' + str(fileset)) if __name__ == '__main__': # Test code/run in file if RUN_THROUGH == FILE: indic_mapping, cmpd_set = extract_drug_indic_mapping(drug_indic_map, min_d=2,key=key, val=val) cando_obj, test_indic, cmpd_to_obj = generate_cando_obj(matrix_file, drug_file, ctd_file, cmpd_set) cando_args = {'cando_obj':cando_obj, 'test_indic':test_indic,\ 'cmpd_to_obj':cmpd_to_obj, 'filename':'0'} bench_args = {'filename':cross_val_file} benchmarking(indic_mapping, cmpd_set, read_strat_xfold_cross, \ FUNC, bench_args, cando_args, out_name + '.tsv') # Command line interface elif RUN_THROUGH == CLI: parser = argparse.ArgumentParser(prog='CANDO Benchmarking', description='Benchmark subsignature scoring methods') # set 1 split to benchmark all indications, 10 to benchmark 1/10th of indications, etc parser.add_argument('num_splits', metavar='number of indication splits', type=str) # 0-indexed, set 0 to assess 1st of num_splits indication subsets & so on parser.add_argument('split_index', metavar='which index split is assessed here', type=int) parser.add_argument('-c', '--ctd', metavar='ctd filepath', type=str) parser.add_argument('-d', '--drug', metavar='drug filepath', type=str) # Note that matrix needs to be scrubbed of cmpd labels # Ie can't use same matrix as subsignature method parser.add_argument('-m', '--matrix', metavar='prot-drug matrix filepath', type=str) # Other values are read from the fileset args = parser.parse_args() # Import indications if args.ctd != None: ctd_file = args.ctd if args.drug != None: drg_file = args.drug #indic_mapping, cmpd_set = import_indications(ctd_file, drug_file) num_splits = int(args.num_splits) split_index = int(args.split_index) standard, cmpd_set = extract_drug_indic_mapping(drug_indic_map, min_d=2) indic_splits = split_indics(standard, num_splits, by='pairs') this_split = indic_splits[split_index] cando_obj, test_indic, cmpd_to_obj = generate_cando_obj(matrix_file, drug_file, ctd_file, cmpd_set) cando_args = {'cando_obj':cando_obj, 'test_indic':test_indic,\ 'cmpd_to_obj':cmpd_to_obj, 'filename':str(split_index)} bench_args = {'filename':cross_val_file} benchmarking(this_split, cmpd_set, read_strat_xfold_cross, \ FUNC, bench_args, cando_args,\ out_name + str(split_index) + '.tsv')