def set_clique_n69(self):
'''
load 69 PCLs currated by Rajiv
Parameters
----------
'''
### load in data for individual groups
llo = ldc.label_loader()
self.pclDict = llo.load_clique_set_n69()
self.all_group_cps = self.pclDict.values()
self.test_groups = self.pclDict.keys()
def set_clique_n69(self):
'''
load 69 PCLs currated by Rajiv
Parameters
----------
'''
### load in data for individual groups
llo = ldc.label_loader()
self.pclDict = llo.load_clique_set_n69()
self.all_group_cps = self.pclDict.values()
self.test_groups = self.pclDict.keys()
def test_classes_incrementally(self, rnkpt_med_file, n_test_max=False):
'''
-start from the most internally consistent PCL - and move down the list
-incrementally increase the number of groups added to the classifier
Parameters
----------
rnkpt_med_file : str
path to a file containing the median summly rankpoint values for each
group (output from the pcla tool)
n_test_max : int
-max number of PCL groups to incorporate into the classifier
-if set to False, all groups are tested
'''
### load in data for individual groups
llo = ldc.label_loader()
self.pclDict = llo.load_TTD()
#load pcl rankpoint file
groupMedians = pd.io.parsers.read_csv(rnkpt_med_file, sep='\t')
groupMedians = groupMedians.sort('median_rankpt', ascending=False)
# make sure compounds are not counted mroe than once in a dictionary:
extendedCompoundList = []
reducedPCLDict = {}
for key in groupMedians['PCL_group']:
value = self.pclDict[key]
for brd in value:
if brd in extendedCompoundList:
value.remove(brd)
reducedPCLDict[key] = value
extendedCompoundList.extend(value)
self.pclDict = reducedPCLDict
# set incrament of groups
if n_test_max:
max_groups = n_test_max
else:
max_groups = groupMedians.shape[0]
group_range = np.arange(2, max_groups + 1)
n_group_accuracy = {}
for n_groups in group_range:
print "testing " + str(n_groups) + " number of classes"
testGroups = groupMedians['PCL_group'][:n_groups].values
self.test_groups = testGroups
self.classification_across_cell(groups_to_model=testGroups,
loo_type='by_cp',
max_signatures_per_cp=3)
n_group_accuracy[n_groups] = self.model_accuracy_across_cells
self.n_group_accuracy = n_group_accuracy
def test_classes_incrementally(self,rnkpt_med_file,n_test_max=False):
'''
-start from the most internally consistent PCL - and move down the list
-incrementally increase the number of groups added to the classifier
Parameters
----------
rnkpt_med_file : str
path to a file containing the median summly rankpoint values for each
group (output from the pcla tool)
n_test_max : int
-max number of PCL groups to incorporate into the classifier
-if set to False, all groups are tested
'''
### load in data for individual groups
llo = ldc.label_loader()
self.pclDict = llo.load_TTD()
#load pcl rankpoint file
groupMedians = pd.io.parsers.read_csv(rnkpt_med_file,sep='\t')
groupMedians = groupMedians.sort('median_rankpt',ascending=False)
# make sure compounds are not counted mroe than once in a dictionary:
extendedCompoundList = []
reducedPCLDict = {}
for key in groupMedians['PCL_group']:
value = self.pclDict[key]
for brd in value:
if brd in extendedCompoundList:
value.remove(brd)
reducedPCLDict[key] = value
extendedCompoundList.extend(value)
self.pclDict = reducedPCLDict
# set incrament of groups
if n_test_max:
max_groups = n_test_max
else:
max_groups = groupMedians.shape[0]
group_range = np.arange(2,max_groups+1)
n_group_accuracy = {}
for n_groups in group_range:
print "testing " + str(n_groups) + " number of classes"
testGroups = groupMedians['PCL_group'][:n_groups].values
self.test_groups = testGroups
self.classification_across_cell(groups_to_model=testGroups,loo_type='by_cp',max_signatures_per_cp=3)
n_group_accuracy[n_groups] = self.model_accuracy_across_cells
self.n_group_accuracy = n_group_accuracy
def set_classes(self):
'''
specify source of class labels
Parameters
----------
'''
### load in data for individual groups
llo = ldc.label_loader()
self.pclDict = llo.load_TTD()
## pick 5 groups - best inter-connectors
testGroups = ['Histone_deacetylase_1-Inhibitor',
'Glucocorticoid_receptor-Agonist',
'Proto-oncogene_tyrosine-protein_kinase_ABL1-Inhibitor',
'Phosphatidylinositol-4,5-bisphosphate_3-kinase_catalytic_subunit,_delta_isoform-Inhibitor',
'3-hydroxy-3-methylglutaryl-coenzyme_A_reductase-Inhibitor']
brdAllGroups = []
for group in testGroups:
brdAllGroups.extend(self.pclDict[group])
self.all_group_cps = brdAllGroups
self.test_groups = testGroups
def set_classes(self):
'''
specify source of class labels
Parameters
----------
'''
### load in data for individual groups
llo = ldc.label_loader()
self.pclDict = llo.load_TTD()
## pick 5 groups - best inter-connectors
testGroups = [
'Histone_deacetylase_1-Inhibitor',
'Glucocorticoid_receptor-Agonist',
'Proto-oncogene_tyrosine-protein_kinase_ABL1-Inhibitor',
'Phosphatidylinositol-4,5-bisphosphate_3-kinase_catalytic_subunit,_delta_isoform-Inhibitor',
'3-hydroxy-3-methylglutaryl-coenzyme_A_reductase-Inhibitor'
]
brdAllGroups = []
for group in testGroups:
brdAllGroups.extend(self.pclDict[group])
self.all_group_cps = brdAllGroups
self.test_groups = testGroups
##
# gmo = gm.GeneMod()
# gmo.load_data_from_gctx(src=cmap.score_path,symbols='TRIB1')
# gmo.z_score_filter(4)
# gmo.signature_strength_filter(8)
# gmo.cell_type_filter('HEPG2')
# # gmo.sc_plot(out=wkdir+'sc_hits.png')
# gmo.scatter()
# gmo.expression_histogram()
# gmo.ssr_plot(out=wkdir+'ssr_hits.png')
# cid = [gmo.cid[x] for x in gmo.reg_ind]
#load in drugbank annotations
reload(ldc)
llo = ldc.label_loader()
pclDict = llo.load_TTD()
dbDict = llo.load_drugbank_by_gene(group_by_action=False)
geneTargets = dbDict.keys()
#put drug-gene relationships in a file
tupList = []
for gene in dbDict:
# make tuple
cps = dbDict[gene]
for cp in cps:
tup = (cp, gene)
tupList.append(tup)
dbFrm = pd.DataFrame(tupList,columns=['brd','target_gene'])
dbSer = pd.Series(dbFrm['target_gene'])
dbSer.index = dbFrm['brd']
cpToGenDict = dbSer.to_dict()
## pick 5 groups - best inter-connectors
# testGroups = ['Histone_deacetylase_1-Inhibitor',
# 'Glucocorticoid_receptor-Agonist',
# 'Proto-oncogene_tyrosine-protein_kinase_ABL1-Inhibitor',
# 'Phosphatidylinositol-4,5-bisphosphate_3-kinase_catalytic_subunit,_delta_isoform-Inhibitor',
# '3-hydroxy-3-methylglutaryl-coenzyme_A_reductase-Inhibitor']
# load in top groups
wkdir = '/xchip/cogs/hogstrom/analysis/scratch'
if not os.path.exists(wkdir):
os.mkdir(wkdir)
#make pso object
pso = psc.svm_pcla(out=wkdir)
self = pso
llo = ldc.label_loader()
self.pclDict = llo.load_TTD()
#load pcl rankpoint file
rnkpt_med_file = '/xchip/cogs/projects/pharm_class/TTd_Oct29/PCL_group_rankpt_medians.txt'
groupMedians = pd.io.parsers.read_csv(rnkpt_med_file, sep='\t')
groupMedians = groupMedians.sort('median_rankpt', ascending=False)
# make sure compounds are not counted mroe than once in a dictionary:
extendedCompoundList = []
reducedPCLDict = {}
for key in groupMedians['PCL_group']:
value = self.pclDict[key]
for brd in value:
if brd in extendedCompoundList:
value.remove(brd)
reducedPCLDict[key] = value
extendedCompoundList.extend(value)