def snpWeights(L, Pi, f, K, t):
n, m = Pi.shape # Dimensions
E = np.empty((n, m), dtype=np.float32)
covariance_cy.updatePCAngsd(L, Pi, E, t)
covariance_cy.standardizeE(E, f, t)
# Performing SVD on normalized expected genotypes
_, s, U = svds(E, k=K)
snpW = U[::-1, :].T * (s[::-1]**
2) / m # Scaling by eigenvalues (PC-scores)
return snpW
def selectionScan(L, Pi, f, K, t):
n, m = Pi.shape # Dimensions
E = np.empty((n, m), dtype=np.float32)
Dsquared = np.empty((m, K), dtype=np.float32)
covariance_cy.updatePCAngsd(L, Pi, E, t)
covariance_cy.standardizeE(E, f, t)
# Performing SVD on normalized expected genotypes
_, _, U = svds(E, k=K)
U = U[::-1, :]
shared.computeD(U, Dsquared)
return Dsquared
def kinshipConomos(L, Pi, t):
n, m = Pi.shape
# Initiate containers
E = np.empty((n, m), dtype=np.float32)
dKin = np.empty(n, dtype=np.float32)
temp1 = np.empty((n, n), dtype=np.float32)
temp2 = np.empty((n, n), dtype=np.float32)
# Dosages
covariance_cy.updatePCAngsd(L, Pi, E, t)
# Kinship computations
kinship_cy.diagKinship(L, Pi, dKin, t)
kinship_cy.numeratorKin(E, Pi, t)
np.dot(E, E.T, out=temp1)
kinship_cy.denominatorKin(E, Pi, t)
np.dot(E, E.T, out=temp2)
temp2 *= 4
temp1 /= temp2
np.fill_diagonal(temp1, dKin) # Insert correct diagonal
return temp1
def pcaEM(L, e, f, m_iter, m_tole, no_std, t):
n, m = L.shape # Dimension of likelihood matrix
n //= 3 # Number of individuals
K = e
# Initiate matrices
E = np.empty((n, m), dtype=np.float32)
dCov = np.zeros(n, dtype=np.float32)
# Estimate covariance matrix (Fumagalli) and infer number of PCs
if K == 0:
# Prepare dosages and diagonal
covariance_cy.covFumagalli(L, f, E, dCov, t)
covariance_cy.standardizeE(E, f, t)
C = np.dot(E, E.T) / m
np.fill_diagonal(C, dCov)
if m_iter == 0:
print("Returning with ngsTools covariance matrix!")
return C, None, K
# Velicer's Minimum Average Partial (MAP) Test
eigVals, eigVecs = eigsh(C, k=min(
n - 1, 15)) # Eigendecomposition (Symmetric) - ARPACK
eigVals = eigVals[::-1] # Sorted eigenvalues
eigVals[eigVals < 0] = 0
eigVecs = eigVecs[:, ::-1] # Sorted eigenvectors
loadings = eigVecs * np.sqrt(eigVals)
mapTest = np.empty(min(m - 1, 15), dtype=np.float32)
# Loop over m-1 eigenvalues for MAP test (Shriner implementation)
for eig in range(min(m - 1, 15)):
partcov = C - (np.dot(loadings[:, 0:(eig + 1)],
loadings[:, 0:(eig + 1)].T))
d = np.diag(partcov)
if (np.sum(np.isnan(d)) > 0) or (np.sum(d == 0) >
0) or (np.sum(d < 0) > 0):
mapTest[eig] = 1
else:
d = np.diagflat(1 / np.sqrt(d))
pr = np.dot(d, np.dot(partcov, d))
mapTest[eig] = (np.sum(pr**2) - m) / (m * (m - 1))
K = max([1, np.argmin(mapTest) + 1
]) # Number of principal components retained
print("Using " + str(K) + " principal components (MAP test)")
# Release memory
del eigVals, eigVecs, loadings, partcov, mapTest
else:
print("Using " + str(K) + " principal components (manually selected)")
covariance_cy.updateFumagalli(L, f, E, t)
# Estimate individual allele frequencies
Pi = np.empty((n, m), dtype=np.float32)
covariance_cy.centerE(E, f, t)
W = estimateSVD(E, K, Pi)
covariance_cy.updatePi(Pi, f, t)
prevW = np.copy(W)
print("Individual allele frequencies estimated (1)")
# Iterative estimation
for iteration in range(2, m_iter + 1):
covariance_cy.updatePCAngsd(L, Pi, E, t)
# Estimate individual allele frequencies
covariance_cy.centerE(E, f, t)
W = estimateSVD(E, K, Pi)
covariance_cy.updatePi(Pi, f, t)
# Break iterative update if converged
diff = covariance_cy.rmse2d_eig(W, prevW)
print("Individual allele frequencies estimated (" + str(iteration) +
"). RMSE=" + str(diff))
if diff < m_tole:
print("Estimation of individual allele frequencies has converged.")
break
prevW = np.copy(W)
del W, prevW
# Estimate covariance matrix (PCAngsd)
if no_std:
covariance_cy.covPCAngsdNoStd(L, f, Pi, E, dCov, t)
covariance_cy.centerE(E, f, t)
else:
covariance_cy.covPCAngsd(L, f, Pi, E, dCov, t)
covariance_cy.standardizeE(E, f, t)
C = np.dot(E, E.T) / m
np.fill_diagonal(C, dCov)
del E
return C, Pi, K
if args.maf_save:
np.save(args.o + ".maf", f.astype(float))
print("Saved population allele frequencies as " + str(args.o) +
".maf.npy (Binary)\n")
if args.indf_save:
np.save(args.o + ".indf", Pi.astype(float))
print("Saved individual allele frequencies as " + str(args.o) +
".indf.npy (Binary)\n")
if args.dosage_save:
import covariance_cy
E = np.empty((L.shape[0] // 3, L.shape[1]),
dtype=np.float32) # Dosage matrix
covariance_cy.updatePCAngsd(L, Pi, E, args.threads)
np.save(args.o + ".dosage", E.astype(float))
print("Saved genotype dosages as " + str(args.o) +
".dosage.npy (Binary)\n")
del E
if args.post_save or args.sites_save:
print("Loading site information")
import pandas as pd
if args.beagle is not None:
infoDF = pd.read_csv(args.beagle,
sep="\t",
header=0,
usecols=[0, 1, 2],
compression="gzip",
dtype=str)