def _getChi2Sum(self, RIDs):
#print([GWAS[x] for x in RIDs])
ps = np.zeros(len(RIDs))
for i in range(0, len(ps)):
#ps = chi2.ppf(1- np.array([GWAS[x] for x in RIDs]),1)
ps[i] = tools.chiSquared1dfInverseCumulativeProbabilityUpperTail(
self._GWAS[RIDs[i]])
return np.sum(ps)
def test_gene_assocdir(self, gene, epsilon=1e-8):
"""
Tests for directional association of the gene
(Requires betas of GWAS)
Args:
gene(str): Gene symbol to test
epsilon(float): Regularization parameter
"""
if gene not in self._GENESYMB:
print(G, "not in annotation!")
return
# Get background
D = self._GENEID[self._GENESYMB[gene]]
REF = {}
REF[D[0]] = self._ref.load_pos_reference(D[0])
if len(self._GWAS_alleles) == 0:
C, RID = self._calcGeneSNPcorr(D[0], self._GENESYMB[gene], REF)
else:
C, RID = self._calcGeneSNPcorr_wAlleles(D[0], self._GENESYMB[gene],
REF)
# Regularize C
A = (C + C.T) / 2
evalu, evec = np.linalg.eigh(A)
evalu[evalu < epsilon] = epsilon
# Decompose
C = evec.dot(np.diag(evalu)).dot(evec.T)
L = np.linalg.cholesky(C)
F = np.linalg.norm(L)**2
# Get and calc stats
z = np.array([
np.sign(self._GWAS_beta[x]) * np.sqrt(
tools.chiSquared1dfInverseCumulativeProbabilityUpperTail(
self._GWAS[x])) for x in RID
])
#print("Right tail:",hpstats.onemin_norm_cdf_100d(np.sum(z),0,F))
#print("Left tail :",hpstats.norm_cdf_100d(np.sum(z),0,F))
return hpstats.onemin_norm_cdf_100d(np.sum(z), 0,
F), hpstats.norm_cdf_100d(
np.sum(z), 0, F)
def _calc_pw_enrichment(self, pw, TISSUES):
RET = {}
FAILS = []
RAW = {}
# Calc pathway enrichment
for tissue in TISSUES:
if tissue not in RAW:
RAW[tissue] = []
A = np.zeros(len(pw))
df = 0
for i in range(0, len(pw)):
if pw[i] in self._GENESYMB:
eid = self._GENESYMB[pw[i]]
pos = np.where(TISSUES[tissue][0] == eid)
if (len(pos[0]) > 0):
A[i] = tools.chiSquared1dfInverseCumulativeProbabilityUpperTail(
TISSUES[tissue][2][pos[0][0]])
df = df + 1
RAW[tissue].append(TISSUES[tissue][2][pos[0][0]])
else:
if pw[i] not in FAILS:
FAILS.append(pw[i])
RAW[tissue].append(np.NaN)
else:
if pw[i] not in FAILS:
FAILS.append(pw[i])
RAW[tissue].append(np.NaN)
if df > 0:
S = np.sum(A)
p = hpstats.onemin_chi2_cdf(S, dof=df)
RET[tissue] = p
else:
RET[tissue] = np.NaN
return RET, FAILS, RAW
def score(self,
modules,
samples=100000,
method='auto',
mode='',
reqacc=1e-100,
threshold=False,
parallel=1,
nobar=False):
"""
Scores a set of pathways/modules
Args:
modules(list): List of modules to score
samples(int): # of random gene sets to draw
method(string): Method to use to evaluate tail probability ('auto','davies','ruben','satterthwaite')
mode(string): Precision mode to use ('','128b','100d')
reqacc(float): requested accuracy
threshold(bool): Threshold p-value to reqacc
nobar(bool): Show progress bar
"""
# Compute fusion sets
if self._fuse:
COMPUTE_SET, FUSION_SET, R = self._genefusion(modules,
method=method,
mode=mode,
reqacc=reqacc,
threshold=threshold,
parallel=parallel,
nobar=nobar)
else:
COMPUTE_SET, FUSION_SET, R = self._nogenefusion(modules)
# Build dictionary
META_DIC = {}
for m in R[0]:
if m[0][:9] == 'METAGENE:':
META_DIC[m[0]] = m[1]
RESULT = []
FAILS = R[1]
# Compute chi2 values for all genes
GENES = {}
for G in self._genescorer._SCORES:
GENES[
G] = tools.chiSquared1dfInverseCumulativeProbabilityUpperTail(
self._genescorer._SCORES[G])
G = list(GENES)
for F in FUSION_SET:
# Compute pathway score
# Calc chi2
chi = np.zeros(len(F[1]))
gpval = np.zeros(len(F[1]))
fail = 0
for i in range(0, len(F[1])):
if F[1][i] in GENES:
chi[i] = GENES[F[1][i]]
gpval[i] = self._genescorer._SCORES[F[1][i]]
else:
fail = fail + 1
gpval[i] = np.NaN
#print("[WARNING]: No gene score for",F[1][i])
L = len(chi) - fail
S = np.sum(chi)
if L > 1:
counter = 0.
B = np.ones(samples)
# Sample background
for i in range(0, samples):
# Draw len(L) random gene-sets
Rs = random.sample(G, L)
b_score = np.zeros(L)
for j in range(0, len(Rs)):
b_score[j] = GENES[Rs[j]]
#print(b_score)
#print(np.sum(b_score))
if np.sum(b_score) > S:
counter += 1.
RESULT.append(
[F[0], F[1], gpval, (1 + counter) / (1 + samples)])
else:
if L == 1:
for j in range(0, len(F[1])):
if F[1][j] in self._genescorer._SCORES:
RESULT.append([
F[0], F[1], chi,
self._genescorer._SCORES[F[1][j]]
])
break
else:
RESULT.append([F[0], F[1], chi, np.NaN])
return [RESULT, FAILS, META_DIC]
def score(self,
modules,
method='auto',
mode='',
reqacc=1e-100,
threshold=False,
parallel=1,
nobar=False):
"""
Scores a set of pathways/modules
Args:
modules(list): List of modules to score
samples(int): # of random gene sets to draw
method(string): Method to use to evaluate tail probability ('auto','davies','ruben','satterthwaite')
mode(string): Precision mode to use ('','128b','100d')
reqacc(float): requested accuracy
threshold(bool): Threshold p-value to reqacc
nobar(bool): Show progress bar
"""
# Compute fusion sets
if self._fuse:
COMPUTE_SET, FUSION_SET, R = self._genefusion(modules,
method=method,
mode=mode,
reqacc=reqacc,
threshold=threshold,
parallel=parallel,
nobar=nobar)
else:
COMPUTE_SET, FUSION_SET, R = self._nogenefusion(modules)
# Build dictionary
META_DIC = {}
for m in R[0]:
if m[0][:9] == 'METAGENE:':
META_DIC[m[0]] = m[1]
# Remove from ._SCORES to have almost same baseline for all modules
for C in COMPUTE_SET:
if C[:9] == 'METAGENE:' and C in self._genescorer._SCORES:
del self._genescorer._SCORES[C]
RESULT = []
FAILS = R[1]
# Score modules
for F in FUSION_SET:
# Note: Ranking inside the F loop because META-GENES have to be added on case by case basis
# Rank gene scores
L = list(self._genescorer._SCORES.keys())
S = []
for i in range(0, len(L)):
S.append(self._genescorer._SCORES[L[i]])
# Add meta genes
for i in range(0, len(F[1])):
if F[1][i][:9] == 'METAGENE:' and F[1][i] in META_DIC:
L.append(F[1][i])
S.append(META_DIC[F[1][i]])
# Remove metagene member genes from background gene list
mgenes = F[1][i][9:].split("_")
for g in mgenes:
if g in self._genescorer._SCORES:
I = L.index(g)
#print(I,"-", g)
del L[I]
del S[I]
# Rank
ra = np.argsort(S)
RANKS = {}
for i in range(0, len(ra)):
RANKS[L[ra[i]]] = (i + 1.) / (len(L) + 1.
) # +1: Ranking t start at 1
# Calc chi2
chi = np.zeros(len(F[1]))
gpval = np.zeros(len(F[1]))
fail = 0
for i in range(0, len(F[1])):
if F[1][i] in RANKS:
chi[i] = tools.chiSquared1dfInverseCumulativeProbabilityUpperTail(
RANKS[F[1][i]])
gpval[i] = RANKS[F[1][i]]
else:
fail = fail + 1
gpval[i] = np.NaN
#print("[WARNING]: No gene score for",F[1][i])
# Calc p-value
df = len(chi) - fail
if df > 0:
S = np.sum(chi)
p = hpstats.onemin_chi2_cdf(S, dof=df)
chi[chi == 0] = np.NaN
RESULT.append([F[0], F[1], gpval, p])
else:
chi[chi == 0] = np.NaN
RESULT.append([F[0], F[1], gpval, np.NaN])
# Cleanup
for G in COMPUTE_SET:
if G in self._genescorer._SCORES:
del self._genescorer._SCORES[G]
# Return
return [RESULT, FAILS, META_DIC]