def main(para):
if 'ListSize' not in para:
para['ListSize'] = '-1'
if 'ThreadNum' not in para:
para['ThreadNum'] = '1'
if 'RandomSeed' not in para:
para['RandomSeed'] = '2014'
if 'ModelName' not in para:
para['ModelName'] = 'RF-bin'
if 'ModelFile' not in para:
para['ModelFile'] = 'set1_all_new.txt.fea.max.RF-bin'
if 'DockMethod' not in para:
para['FeatureType'] = 'SaveResidue'
if 'SolutionNum' not in para:
para['SolutionNum'] = '10'
if 'PredictCutoff' not in para:
para['PredictCutoff'] = '0.5'
## Step 1: docking all pdb chain pairs and train
if not os.path.exists(para['ModelFile']):
import cross_validation
para1 = para.copy()
para1['SplitFold'] = '1'
cross_validation.main(para1)
## Step 2: docking new pairs and predict
## prepare pdb pair list
pdblistfile = 'list_from_user.txt'
with open(pdblistfile, 'w') as tempfile:
tempfile.write('Hhp1\tTas3\t4HOK\tA\t3D1D\tA\n')
#tempfile.write('Hhp1\tMoc3\t4HOK\tA\tMOC3_modbase\t \n')
tempfile.write('Hhp1\tPpc89\t4HOK\tA\tPPC89_Modbase\t \n')
## docking them and generate features
feature_file = 'features_for_predicting.txt'
#if not os.path.exists(feature_file):
if True:
import prepare_feature
para2 = para.copy()
para2['ListFile'] = pdblistfile
para2['OutFile'] = feature_file
para2['ListFormat'] = 'p1/p2/pdb1/ch1/pdb2/ch2'
prepare_feature.main(para2)
## 2.c: predict their contact probabilities
from cross_validation import add_residue_label, model_predict
add_residue_label(feature_file)
predfile = model_predict(feature_file, model=para['ModelName'], mfile=para['ModelFile'])
## Step 3: get predicted values for each residue
from cross_validation import map_pdb_residue
residue_value = map_pdb_residue(predfile)
with open(para['ExeFile']+'data.txt', 'w') as outfile:
for g1g2, res, val in residue_value:
outfile.write('%s\t%s\t-1\t%s\n'%('\t'.join(g1g2.split('=')), res, val))
from evaluate_perform import read_residue_data, group_residue
idx, val1, val2 = read_residue_data(para['ExeFile']+'data.txt')
pp_val = group_residue(idx, val2) ## using predicted value
for p1, p2 in pp_val:
res = pp_val[(p1,p2)]
pp = list(set([r.split(':')[0] for r in res]))
if True: ## reformat protein names
if p1 == p2 and len(pp) == 1:
p1 = pp
p2 = pp
elif len(pp) == 2:
p1 = pp[0]
p2 = pp[1]
else:
print 'Failed to map', p1, p2, pp
show(p1)
show(p2)
int1 = [r for r in res if r.split(':')[0]==p1 and res[r] >= float(para['PredictCutoff'])]
int2 = [r for r in res if r.split(':')[0]==p2 and res[r] >= float(para['PredictCutoff'])]
ord1 = sorted([int(r.split(':')[-1]) for r in int1])
ord2 = sorted([int(r.split(':')[-1]) for r in int2])
show(','.join([str(i) for i in ord1]))
show(','.join([str(i) for i in ord2]))
show()
def main(para):
if 'ListFile' not in para:
para['ListFile'] = para['DataPath']+'/set1_all_new.txt'
if 'SplitFold' not in para:
para['SplitFold'] = '10'
if 'SolutionNum' not in para:
para['SolutionNum'] = '10'
if 'RandomSeed' not in para:
para['RandomSeed'] = '2014'
if 'ThreadNum' not in para:
para['ThreadNum'] = '1'
if 'FeatureType' not in para:
para['FeatureType'] = 'SaveResidue'
if 'ModelName' not in para:
para['ModelName'] = 'RF-bin'
train_list = [] ## unique pdb list for training
real_value = [] ## from cocrystal
pred_value = [] ## from learning model
other_vals = {}
if para['FeatureType'] == 'SaveResidue':
other_vals = {2:[], 3:[], 4:[], 5:[]}
# elif para['FeatureType'] == 'SaveZDOCK':
# other_vals = {2:[], 3:[], 4:[]}
elif para['FeatureType'] == 'SavePatchDock':
other_vals = {2:[], 3:[], 4:[]}
elif para['FeatureType'] == 'SaveSequence':
other_vals = {2:[], 3:[], 4:[]}
all_res, pp_int = get_res_labels(para)
import prepare_feature
for train, test in split_data(para['ListFile'], int(para['SplitFold']), int(para['RandomSeed'])):
if para['SplitFold'] == '1':
train = para['ListFile']
test = para['ListFile']
if False: ## compare DDI network
from generate_hSIN import generate_ddi2, get_pdb_subset
generate_ddi2(output_file = train + '.ddi',
pdb_subset = get_pdb_subset(train))
generate_ddi2(output_file = test + '.ddi',
pdb_subset = get_pdb_subset(test))
from domain_map import reduced_ddi
train_ddi = reduced_ddi(train + '.ddi')
test_ddi = reduced_ddi(test + '.ddi')
###############################################################
## Train
show(train, False)
para2 = para.copy() ## copy parameters
para2['ListFile'] = train
para2['ListSize'] = '-1'
para2['ListFormat'] = 'p1/p2/pdb1/ch1/pdb2/ch2'
prepare_feature.main(para2)
resfile = combine_pdb_residue(para2['OutFile'])
#resfile = add_more_info(resfile, train_ddi)
show(add_residue_label(resfile, pp_int), False)
mfile = train_model(resfile, model=para['ModelName'])
###############################################################
## Test
show(test, False)
para2['ListFile'] = test
prepare_feature.main(para2)
resfile = combine_pdb_residue(para2['OutFile'])
#resfile = add_more_info(resfile, train_ddi)
show(add_residue_label(resfile, None), False)
outfile = model_predict(resfile, model=para['ModelName'], mfile=mfile)
## Save values
if not os.path.exists(outfile):
continue ## skip this fold
pred_value += map_pdb_residue(outfile, para2['MapFile'])
for idx in other_vals:
values = other_vals[idx]
values += map_pdb_residue(resfile, para2['MapFile'], useidx=idx)
other_vals[idx] = values
show()
if para['SplitFold'] != '1':
## clean files
os.system('rm %s*'%train)
os.system('rm %s*'%test)
## Get the real labels of residues in predicted protein pairs
pred_pp = set([pp for pp, res, val in pred_value])
show('Performance based on %s protein pairs'%len(pred_pp))
for pp, res, val in all_res:
if pp in pred_pp:
real_value.append((pp, res, val))
## Comparison between know_value and pred_value agaist real_value
save_list = [('cocry', real_value), ('pred'+'-'+para['FeatureType']+'-'+para['ModelName'], pred_value)]
for idx in other_vals:
save_list.append((para['FeatureType']+'-idx%s'%idx, other_vals[idx]))
for name, value in save_list:
head = []; real = []; pred = [];
for idx, res, val1, val2 in intersect_poslist(real_value, value):
p1, p2 = idx.split('=')
head.append([p1, p2, res])
real.append(val1)
pred.append(val2)
with open('cv_cocry_'+name+'_'+para['RandomSeed']+'.txt', 'w') as tempfile:
for _a, _b, _c in zip(head, real, pred):
tempfile.write('%s\t%s\t%s\n'%('\t'.join(_a), _b, _c))
## evaluate
show(name, False)
area, px, py, pc = performance(real, pred, x='FPR', y='TPR')
show(area, False)
#show(); show('FPR'); show(px, True);
#show('TPR'); show(py, True);
area, px, py, pc = performance(real, pred, x='TPR', y='PPV')
show(area, False)
#show(); show('Recall'); show(px, True);
#show('Precision'); show(py, True);
#show('Cutoff'); show(pc, True);
show()
