def make_bkb(self, train_patients):
t1 = time.time()
#-- Make Fused BKB with training paitents
#-- Collect all patient BKFs
bkfs = list()
hashes = list()
for patient_hash in tqdm.tqdm(train_patients,
leave=False,
desc='Preparing BKFs for Fusion'):
bkf_file = os.path.join(self.bkf_folder,
'{}.bkf'.format(patient_hash))
_bkf = BKB()
try:
_bkf.load(bkf_file)
bkfs.append(_bkf)
hashes.append(str(patient_hash))
except:
continue
#-- Try to add PatientX
try:
bkf_file = os.path.join(self.bkf_folder, 'PatientX.bkf')
_bkf = BKB()
_bkf.load(bkf_file)
bkfs.append(_bkf)
hashes.append('PatientX')
except:
pass
fused_bkb = fuse(bkfs, [1 for _ in range(len(bkfs))], hashes)
logging.info('Make BKB Ok.')
logging.info('Elapsed Time: {} sec'.format(time.time() - t1))
return fused_bkb
def __init__(self, config_file):
#-- Read in configuration from config file.
self.config = dict()
with open(config_file, 'r') as csv_file:
reader = csv.reader(csv_file)
for row in reader:
self.config[row[0]] = row[1]
self.fused_bkb_path = self.config['fused_bkb_path']
self.ncats_dir = self.config['ncats_dir']
self.src_metadata_path = self.config['src_metadata_path']
self.gene_var_direct = self.config['gene_var_direct']
self.fused_bkb = BKB()
self.fused_bkb.load(self.fused_bkb_path)
self.reasoner = Reasoner(self.fused_bkb,
gene_var_direct=self.gene_var_direct,
max_new_ev=MAX_NEW_EV)
self.reasoner.set_src_metadata(self.src_metadata_path)
def processPathwayBKF(self, exhaustiveOr=False):
assert len(self.pathways) > 0, "Have not processed pathways yet"
for pathway in tqdm.tqdm(self.pathways,
desc='Processing pathway BKFs'):
bkf = BKB(name=pathway.pathwayID)
pathwayActiveComp_idx = bkf.addComponent('{}_active='.format(
pathway.pathwayID))
pathwayActiveTrue_idx = bkf.addComponentState(
pathwayActiveComp_idx, 'True')
if exhaustiveOr:
print("not implemented")
else:
geneSelectorComp_idx = bkf.addComponent("Gene_combo=")
for gene in pathway.genes: #gene[0] = geneName, gene[1] = low value gene[2] = high value
geneCombo_idx = bkf.addComponentState(
geneSelectorComp_idx, gene[0])
statConditionComp_idx = bkf.addComponent(
'mu-STD<={}<=mu+STD='.format(gene[0]))
statConditionTrue_idx = bkf.addComponentState(
statConditionComp_idx, 'True')
bkf.addSNode(
BKB_S_node(statConditionComp_idx,
statConditionTrue_idx, 1.0))
bkf.addSNode(
BKB_S_node(
geneSelectorComp_idx, geneCombo_idx, 1.0,
[(statConditionComp_idx, statConditionTrue_idx)]))
bkf.addSNode(
BKB_S_node(pathwayActiveComp_idx,
pathwayActiveTrue_idx, 1.0,
[(geneSelectorComp_idx, geneCombo_idx)]))
self.bkfs.append(bkf)
assert len(self.pathways) > 0, "Have not processed pathways yet"
def processPathwayBKF(self, exhaustiveOr=False):
assert len(self.pathways) > 0, "Have not processed pathways yet"
for pathway in tqdm.tqdm(self.pathways,
desc='Processing pathway BKFs'):
bkf = BKB(name=pathway.ID)
if "_hier" in pathway.ID:
#head
pathwayParentComp_idx = bkf.addComponent('{}_active='.format(
pathway.head))
pathwayParentTrue_idx = bkf.addComponentState(
pathwayParentComp_idx, 'True')
#tail - should only be 1 for _hier bkfs
pathwayChildComp_idx = bkf.addComponent('{}_active='.format(
pathway.tails[0]))
pathwayChildTrue_idx = bkf.addComponentState(
pathwayChildComp_idx, 'True')
#S-node
bkf.addSNode(
BKB_S_node(pathwayChildComp_idx, pathwayChildTrue_idx,
1.0))
bkf.addSNode(
BKB_S_node(pathwayParentComp_idx, pathwayParentTrue_idx,
1.0,
[(pathwayChildComp_idx, pathwayChildTrue_idx)]))
else:
'''
#head
pathwayHierComp_idx = bkf.addComponent('{}_active='.format(pathway.head))
pathwayHierTrue_idx = bkf.addComponentState(pathwayHierComp_idx, 'True')
#tails
for tail in pathway.tails:
statConditionComp_idx = bkf.addComponent('mu-STD<={}<=mu+STD'.format(tail))
statConditionTrue_idx = bkf.addComponentState(statConditionComp_idx, 'True')
bkf.addSNode(BKB_S_node(statConditionComp_idx, statConditionTrue_idx, 1.0))
bkf.addSNode(BKB_S_node(pathwayHierComp_idx, pathwayHierTrue_idx, 1.0, [(statConditionComp_idx, statConditionTrue_idx)]))
self.bkfs.append(bkf)
'''
#=======
if exhaustiveOr:
print("not implemented")
else:
# pathway
pathwayReactionComp_idx = bkf.addComponent(pathway.head +
"_active=")
pathwayReactionTrue_idx = bkf.addComponentState(
pathwayReactionComp_idx, 'True')
# gene slector
geneSelectorComp_idx = bkf.addComponent("Gene_combo=")
#tails
for tail in pathway.tails:
geneCombo_idx = bkf.addComponentState(
geneSelectorComp_idx, tail)
statConditionComp = BKB_component("mu-STD>=" + tail +
"<=mu+STD=")
statConditionTrue = BKB_I_node('True',
statConditionComp)
statConditionComp_idx = bkf.addComponent("mu-STD>=" +
tail +
"<=mu+STD=")
statConditionTrue_idx = bkf.addComponentState(
statConditionComp_idx, 'True')
bkf.addSNode(
BKB_S_node(statConditionComp_idx,
statConditionTrue_idx, 1.0))
bkf.addSNode(
BKB_S_node(geneSelectorComp_idx, geneCombo_idx,
1.0, [(statConditionComp_idx,
statConditionTrue_idx)]))
bkf.addSNode(
BKB_S_node(
pathwayReactionComp_idx,
pathwayReactionTrue_idx, 1.0,
[(geneSelectorComp_idx, geneCombo_idx)]))
self.bkfs.append(bkf)
Thayer School of Engineering at Dartmouth College
Authors: Dr. Eugene Santos, Jr
Mr. Chase Yakaboski,
Mr. Gregory Hyde,
Dr. Keum Joo Kim
'''
#-- This test case took 35,000 seconds = 9.7 hours
#-- Import your version of BKB and reasoning appropriately. Below is how I do it with pybkb.
from pybkb.common.bayesianKnowledgeBase import bayesianKnowledgeBase as BKB
from pybkb.python_base.reasoning import updating
#-- Other imports
import pickle
#-- Load in evidence and targets
with open('query_evid_targs.pk', 'rb') as f_:
data = pickle.load(f_)
#-- Pickled as dictionary.
evidence = data['evidence']
targets = data['targets']
#-- Load query BKB
bkb = BKB()
bkb.load('query_bkb_137.bkb')
#-- Run Updating
res = updating(bkb, evidence, targets)
import sys
from pybkb.common.bayesianKnowledgeBase import bayesianKnowledgeBase as BKB
from pybkb.python_base.reasoning import checkMutex
#!!!!!!!!!! Change to your local bkb-pathway-core directory.
#sys.path.append('/home/ghyde/bkb-pathway-provider/core')
#sys.path.append('/home/cyakaboski/src/python/projects/bkb-pathway-provider/core')
from chp.reasoner import Reasoner
from chp.query import Query
#-- Initalize a BKB
fused_bkb = BKB()
'''
#-- Load in the fused bkb from our datafiles
fused_bkb.load('/home/public/data/ncats/AxleBKBS/660Pats6HoldoutSTAGING/fusion.bkb')
#-- Here are the associated patient data files
patient_data_file = '/home/public/data/ncats/AxleBKBS/660Pats6HoldoutSTAGING/patient_data.pk'
withheld_patients_file = '/home/public/data/ncats/AxleBKBS/660Pats6HoldoutSTAGING/withheldPatients.csv'
'''
#-- Load in the fused bkb from our datafiles
fused_bkb.load('/home/public/data/ncats/BabelBKBs/smallProblem/fusion.bkb')
#-- Here are the associated patient data files
patient_data_file = '/home/public/data/ncats/BabelBKBs/smallProblem/patient_data.pk'
withheld_patients_file = '/home/public/data/ncats/BabelBKBs/smallProblem/withheldPatients.csv'
#-- Instiante reasoner
reasoner = Reasoner(fused_bkb=fused_bkb)
def read_expression_snodes(self, expression_levels_dict,
tcga_expression_file, react_gene_file,
pathways_file):
#-- Process all gene_reaction s_nodes and assume there is a header with gene names.
with open(react_gene_file, 'r') as csv_file:
reader = csv.reader(csv_file)
rows = [row for row in reader]
#-- Process Gene Names from header:
print(rows[0])
genes = rows[0][1:]
genes = [genes[i][:-4] for i in range(0, len(genes), 2)]
print(genes)
for row in rows[1:]:
bkf = BKB(name=row[0] + ' BKF')
reaction_head_name = row[0]
react_component = BKB_component(reaction_head_name)
i_node_true = BKB_I_node('True', react_component)
i_node_false = BKB_I_node('False', react_component)
react_component.addINode(i_node_true)
react_component.addINode(i_node_false)
bkf.addComponent(react_component)
#-- Extract and Make I-nodes
for gene_j, row_i in enumerate(range(1, len(row), 2)):
gene_name = genes[gene_j]
gene_low = row[row_i]
gene_high = row[row_i + 1]
if gene_low != '':
print(gene_low, gene_high)
num_levels = expression_levels_dict[gene_name]
levels = [
rag for rag in np.linspace(int(gene_low), int(
gene_high), num_levels)
]
component_states = [
'{} | {}'.format(levels[i], levels[i + 1])
for i in range(len(levels) - 1)
]
component = BKB_component(gene_name + ' ExpressionLv')
for state in component_states:
i_node = BKB_I_node(state, component)
component.addINode(i_node)
bkf.addComponent(component)
for idx in range(
bkf.findComponent(
component.name).getNumberStates()):
i_node = component.getState(idx)
bkf.addSNode(
BKB_S_node(component, i_node, random.random()))
bkf.addSNode(
BKB_S_node(react_component, i_node_true,
random.random(), [(component, i_node)]))
bkf.addSNode(
BKB_S_node(react_component, i_node_false,
random.random(), [(component, i_node)]))
self.bkfs.append(bkf)
print('Done')
#-- Process Pathway BKFs
with open(pathways_file, 'r') as csv_file:
reader = csv.reader(csv_file)
bkf = BKB()
for row in reader:
tail = row[0]
head = row[1]
states = ['True', 'False']
if bkf.findComponent(tail) == -1:
component_tail = BKB_component(tail)
for state in states:
component_tail.addINode(
BKB_I_node(state, component_tail))
bkf.addComponent(component_tail)
else:
component_tail = bkf.findComponent(tail)
if bkf.findComponent(head) == -1:
component_head = BKB_component(head)
for state in states:
component_head.addINode(
BKB_I_node(state, component_head))
bkf.addComponent(component_head)
else:
component_head = bkf.findComponent(head)
for state1 in states:
for state2 in states:
bkf.addSNode(
BKB_S_node(component_head,
component_head.findState(state1),
random.random(),
[(component_tail,
component_tail.findState(state2))]))
self.bkfs.append(bkf)
print('Complete')