def calculate_ld_tables(input_genotype_file, chrom_i, local_ld_hdf5_file, ld_radius,
min_r2=0.2, maf_thres=0.01, indiv_filter=None,
snp_filter=None, return_void=True, verbose=True):
"""
Calculate the LD tables for the given radius, and store in the given file.
"""
if not os.path.isfile(local_ld_hdf5_file):
h5f = h5py.File(input_genotype_file)
print 'Calculating LD information for chromosome %d w. radius %d' % (chrom_i, ld_radius)
g_dict = kgenome.get_genotype_data(h5f, chrom_i, maf_thres, indiv_filter=indiv_filter,
snp_filter=snp_filter, randomize_sign=False, snps_signs=None)
ld_dict = get_ld_table(g_dict['norm_snps'], ld_radius=ld_radius, min_r2=min_r2, verbose=verbose)
ld_dict['avg_ld_score'] = sp.mean(ld_dict['ld_scores'])
print 'Done calculating the LD table and LD score, writing to file:', local_ld_hdf5_file
oh5f = h5py.File(local_ld_hdf5_file, 'w')
hu.dict_to_hdf5(ld_dict, oh5f)
oh5f.close()
print 'LD information is now stored.'
else:
print 'Loading LD information from file: %s' % local_ld_hdf5_file
oh5f = h5py.File(local_ld_hdf5_file, 'r')
ld_dict = hu.hdf5_to_dict(oh5f)
oh5f.close()
if not return_void:
return ld_dict
def calculate_ld_tables(input_genotype_file,
chrom_i,
local_ld_hdf5_file,
ld_radius,
min_r2=0.2,
maf_thres=0.01,
indiv_filter=None,
snp_filter=None,
return_void=True,
verbose=True):
"""
Calculate the LD tables for the given radius, and store in the given file.
"""
if not os.path.isfile(local_ld_hdf5_file):
h5f = h5py.File(input_genotype_file)
print 'Calculating LD information for chromosome %d w. radius %d' % (
chrom_i, ld_radius)
g_dict = kgenome.get_genotype_data(h5f,
chrom_i,
maf_thres,
indiv_filter=indiv_filter,
snp_filter=snp_filter,
randomize_sign=False,
snps_signs=None)
ld_dict = get_ld_table(g_dict['norm_snps'],
ld_radius=ld_radius,
min_r2=min_r2,
verbose=verbose)
ld_dict['avg_ld_score'] = sp.mean(ld_dict['ld_scores'])
print 'Done calculating the LD table and LD score, writing to file:', local_ld_hdf5_file
oh5f = h5py.File(local_ld_hdf5_file, 'w')
hu.dict_to_hdf5(ld_dict, oh5f)
oh5f.close()
print 'LD information is now stored.'
else:
print 'Loading LD information from file: %s' % local_ld_hdf5_file
oh5f = h5py.File(local_ld_hdf5_file, 'r')
ld_dict = hu.hdf5_to_dict(oh5f)
oh5f.close()
if not return_void:
return ld_dict
def calculate(
input_genotype_file,
input_ld_pruned_genotype_file,
ld_score_file,
kinship_pca_file,
ld_radius=200,
maf_thres=0.01,
snp_filter_frac=0.05,
debug_filter_frac=1,
):
"""
Generates population structure adjusted 1k genomes LD scores and stores in the given file.
"""
# Kinship
kinship_pca_dict = kgenome.get_kinship_pca_dict(
input_ld_pruned_genotype_file, kinship_pca_file, maf_thres=maf_thres, snp_filter_frac=snp_filter_frac
)
chrom_snp_trans_mats = {}
for chrom in range(1, 23):
print "Working on Chromosome %d" % chrom
chrom_str = "chr%d" % chrom
chrom_snp_trans_mats[chrom_str] = kinship_pca_dict[chrom_str]["cholesky_decomp_inv_snp_cov"]
# bla
lds_dict = generate_1k_LD_scores(
input_genotype_file,
chrom_snp_trans_mats,
maf_thres=maf_thres,
ld_radius=ld_radius,
debug_filter_frac=debug_filter_frac,
)
# Store LD scores
lds_h5f = h5py.File(ld_score_file)
hu.dict_to_hdf5(lds_dict, lds_h5f)
lds_h5f.close()
def calculate(input_genotype_file,
input_ld_pruned_genotype_file,
ld_score_file,
kinship_pca_file,
ld_radius=200,
maf_thres=0.01,
snp_filter_frac=0.05,
debug_filter_frac=1):
"""
Generates population structure adjusted 1k genomes LD scores and stores in the given file.
"""
# Kinship
kinship_pca_dict = kgenome.get_kinship_pca_dict(
input_ld_pruned_genotype_file,
kinship_pca_file,
maf_thres=maf_thres,
snp_filter_frac=snp_filter_frac)
chrom_snp_trans_mats = {}
for chrom in range(1, 23):
print 'Working on Chromosome %d' % chrom
chrom_str = 'chr%d' % chrom
chrom_snp_trans_mats[chrom_str] = kinship_pca_dict[chrom_str][
'cholesky_decomp_inv_snp_cov']
# bla
lds_dict = generate_1k_LD_scores(input_genotype_file,
chrom_snp_trans_mats,
maf_thres=maf_thres,
ld_radius=ld_radius,
debug_filter_frac=debug_filter_frac)
# Store LD scores
lds_h5f = h5py.File(ld_score_file)
hu.dict_to_hdf5(lds_dict, lds_h5f)
lds_h5f.close()
def calc_kinship(input_file='Data/1Kgenomes/1K_genomes_v3.hdf5',
out_file='Data/1Kgenomes/kinship.hdf5',
maf_thres=0.01,
figure_dir='',
figure_fn='',
snp_filter_frac=1,
indiv_filter_frac=1,
chrom_ok_snp_dict=None):
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
print 'Loading Genotype from '
in_h5f = h5py.File(input_file)
# eur_filter = in_h5f['indivs']['continent'][...] == 'EUR'
# num_indivs = sp.sum(eur_filter)
indiv_ids = in_h5f['indiv_ids'][...]
indiv_filter = None
if indiv_filter_frac < 1:
indiv_filter = sp.array(
sp.random.random(len(indiv_ids)) < indiv_filter_frac,
dtype='bool8')
indiv_ids = indiv_ids[indiv_filter]
assert len(sp.unique(indiv_ids)) == len(indiv_ids)
num_indivs = len(indiv_ids)
ok_chromosome_dict = {}
not_done = set(range(1, 23))
while len(not_done) > 0:
chromosome_dict = {}
K_all_snps = sp.zeros((num_indivs, num_indivs), dtype='float32')
num_all_snps = 0
sum_indiv_genotypes_all_chrom = sp.zeros(num_indivs, dtype='float32')
# snp_cov_all_snps = sp.zeros((num_indivs, num_indivs), dtype='float64')
print 'Calculating kinship'
for chrom in range(1, 23):
print 'Working on Chromosome %d' % chrom
chrom_str = 'chr%d' % chrom
snp_filter = None
if snp_filter_frac < 1:
snp_filter = sp.random.random(len(
in_h5f[chrom_str]['snps'])) < snp_filter_frac
g_dict = get_genotype_data(in_h5f,
chrom,
maf_thres,
indiv_filter=indiv_filter,
snp_filter=snp_filter,
randomize_sign=True,
snps_signs=None,
chrom_ok_snp_dict=chrom_ok_snp_dict)
norm_snps = g_dict['norm_snps']
sum_indiv_genotypes = sp.sum(g_dict['norm_snps'], 0)
sum_indiv_genotypes_all_chrom += sum_indiv_genotypes
print 'Calculating chromosome kinship'
K_unscaled = sp.array(sp.dot(norm_snps.T, norm_snps),
dtype='float32')
assert sp.isclose(
sp.sum(sp.diag(K_unscaled)) / (len(norm_snps) * num_indivs),
1.0), '..bug'
K_all_snps += K_unscaled
num_all_snps += len(norm_snps)
print 'SNP-cov normalisation'
sum_indiv_genotypes = sp.sum(norm_snps, 0)
sum_indiv_genotypes_all_chrom += sum_indiv_genotypes
mean_indiv_genotypes = sum_indiv_genotypes / len(norm_snps)
norm_snps = norm_snps - mean_indiv_genotypes
print 'Calculating SNP covariance unscaled'
snp_cov_unscaled = sp.array(sp.dot(norm_snps.T, norm_snps),
dtype='float32')
# snp_cov_all_snps += snp_cov_unscaled
print 'Storing and updating things'
chromosome_dict[chrom_str] = {
'K_unscaled': K_unscaled,
'num_snps': len(norm_snps),
'sum_indiv_genotypes': sum_indiv_genotypes,
'snp_cov_unscaled': snp_cov_unscaled,
'snps_signs': g_dict['snps_signs']
}
if snp_filter_frac < 1:
chromosome_dict[chrom_str]['snp_filter'] = snp_filter
# snp_cov_all_snps = snp_cov_all_snps / float(num_all_snps)
# K_all_snps = K_all_snps / float(num_all_snps)
# print 'K_all_snps.shape: %s' % str(K_all_snps.shape)
# print 'snp_cov_all_snps.shape: %s' % str(snp_cov_all_snps.shape)
# print 'sp.diag(snp_cov_all_snps): %s' % str(sp.diag(snp_cov_all_snps))
# print 'sp.mean(sp.diag(snp_cov_all_snps)_: %s' % str(sp.mean(sp.diag(snp_cov_all_snps)))
# print 'Full kinship and snp-covariance calculation done using %d SNPs\n' % num_all_snps
mean_indiv_genotypes_all_chrom = sum_indiv_genotypes_all_chrom / num_all_snps
print 'Individual gentoype mean found:'
print mean_indiv_genotypes_all_chrom
print 'Calculating chromosome-wise SNP-covariance and kinship matrices'
for chrom in range(1, 23):
if chrom in not_done:
print 'Working on Chromosome %d' % chrom
chrom_str = 'chr%d' % chrom
snp_cov_leave_one_out = sp.zeros((num_indivs, num_indivs),
dtype='float32')
K_leave_one_out = sp.zeros((num_indivs, num_indivs),
dtype='float32')
num_snps_used = 0
sum_indiv_genotypes = sp.zeros(num_indivs, dtype='float32')
for chrom2 in range(1, 23):
chrom2_str = 'chr%d' % chrom2
if chrom2 != chrom:
sum_indiv_genotypes += chromosome_dict[chrom2_str][
'sum_indiv_genotypes']
K_leave_one_out += chromosome_dict[chrom2_str][
'K_unscaled']
num_snps_used += chromosome_dict[chrom2_str][
'num_snps']
assert sp.isclose(
sp.sum(sp.diag(K_leave_one_out)) /
(num_snps_used * num_indivs), 1.0), '..bug'
mean_indiv_genotypes = sum_indiv_genotypes / num_snps_used
for chrom2 in range(1, 23):
chrom2_str = 'chr%d' % chrom2
if chrom2 != chrom:
print 'Loading SNPs'
snps_signs = chromosome_dict[chrom2_str]['snps_signs']
snp_filter = chromosome_dict[chrom2_str]['snp_filter']
g_dict = get_genotype_data(
in_h5f,
chrom2,
maf_thres,
indiv_filter=indiv_filter,
snp_filter=snp_filter,
randomize_sign=True,
snps_signs=snps_signs,
chrom_ok_snp_dict=chrom_ok_snp_dict)
norm_snps = g_dict['norm_snps']
print 'SNP-cov normalisation'
norm_snps = norm_snps - mean_indiv_genotypes
print 'Calculating SNP covariance unscaled'
snp_cov_unscaled = sp.dot(norm_snps.T, norm_snps)
snp_cov_leave_one_out += snp_cov_unscaled
snp_cov_leave_one_out = snp_cov_leave_one_out / num_snps_used
K_leave_one_out = K_leave_one_out / num_snps_used
assert (K_leave_one_out -
sp.diag(K_leave_one_out)).max() < 0.1, '..bug'
try:
cholesky_decomp_inv_snp_cov = linalg.cholesky(
linalg.pinv(
sp.array(snp_cov_leave_one_out, dtype='float64')))
evals, evecs = linalg.eig(
sp.array(K_leave_one_out, dtype='float64'))
except:
try:
cholesky_decomp_inv_snp_cov = linalg.cholesky(
linalg.pinv(
sp.array(snp_cov_leave_one_out,
dtype='float32')))
evals, evecs = linalg.eig(
sp.array(K_leave_one_out, dtype='float32'))
except:
print 'Failed when obtaining the Cholesky decomposotion or eigen decomposition'
print 'Moving on, trying again later.'
continue
sort_indices = sp.argsort(evals, )
ordered_evals = evals[sort_indices]
print ordered_evals[-10:] / sp.sum(ordered_evals)
ordered_evecs = evecs[:, sort_indices]
d = {}
d['evecs_leave_one_out'] = ordered_evecs
d['evals_leave_one_out'] = ordered_evals
d['cholesky_decomp_inv_snp_cov'] = cholesky_decomp_inv_snp_cov
d['K_leave_one_out'] = K_leave_one_out
d['K_unscaled'] = chromosome_dict[chrom_str]['K_unscaled']
d['num_snps'] = chromosome_dict[chrom_str]['num_snps']
d['snp_cov_leave_one_out'] = snp_cov_leave_one_out
ok_chromosome_dict[chrom_str] = d
not_done.remove(chrom)
# While loop ends here.
K_all_snps = K_all_snps / float(num_all_snps)
in_h5f.close()
ok_chromosome_dict['K_all_snps'] = K_all_snps
ok_chromosome_dict['num_all_snps'] = num_all_snps
assert sp.sum((ok_chromosome_dict['chr1']['K_leave_one_out'] -
ok_chromosome_dict['chr2']['K_leave_one_out'])**
2) != 0, 'Kinships are probably too similar.'
print 'Calculating PCAs'
evals, evecs = linalg.eigh(sp.array(
K_all_snps, dtype='float64')) # PCA via eigen decomp
evals[evals < 0] = 0
sort_indices = sp.argsort(evals, )[::-1]
ordered_evals = evals[sort_indices]
print ordered_evals[:10] / sp.sum(ordered_evals)
pcs = evecs[:, sort_indices]
tot = sum(evals)
var_exp = [(i / tot) * 100 for i in sorted(evals, reverse=True)]
print 'Total variance explained:', sp.sum(var_exp)
ok_chromosome_dict['pcs'] = pcs
ok_chromosome_dict['pcs_var_exp'] = var_exp
if figure_dir is not None:
plt.clf()
plt.plot(pcs[:, 0], pcs[:, 1], 'k.')
plt.title("Overall PCA")
plt.xlabel('PC1')
plt.xlabel('PC2')
plt.tight_layout()
plt.savefig(figure_dir + '/' + figure_fn, format='pdf')
plt.clf()
out_h5f = h5py.File(out_file)
hu.dict_to_hdf5(ok_chromosome_dict, out_h5f)
out_h5f.close()
return ok_chromosome_dict
def calc_kinship(input_file='Data/1Kgenomes/1K_genomes_v3.hdf5' , out_file='Data/1Kgenomes/kinship.hdf5',
maf_thres=0.01, figure_dir='', figure_fn='', snp_filter_frac=1, indiv_filter_frac=1,
chrom_ok_snp_dict=None):
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
print 'Loading Genotype from '
in_h5f = h5py.File(input_file)
# eur_filter = in_h5f['indivs']['continent'][...] == 'EUR'
# num_indivs = sp.sum(eur_filter)
indiv_ids = in_h5f['indiv_ids'][...]
indiv_filter = None
if indiv_filter_frac < 1:
indiv_filter = sp.array(sp.random.random(len(indiv_ids)) < indiv_filter_frac, dtype='bool8')
indiv_ids = indiv_ids[indiv_filter]
assert len(sp.unique(indiv_ids)) == len(indiv_ids)
num_indivs = len(indiv_ids)
ok_chromosome_dict = {}
not_done = set(range(1, 23))
while len(not_done) > 0:
chromosome_dict = {}
K_all_snps = sp.zeros((num_indivs, num_indivs), dtype='float32')
num_all_snps = 0
sum_indiv_genotypes_all_chrom = sp.zeros(num_indivs, dtype='float32')
# snp_cov_all_snps = sp.zeros((num_indivs, num_indivs), dtype='float64')
print 'Calculating kinship'
for chrom in range(1, 23):
print 'Working on Chromosome %d' % chrom
chrom_str = 'chr%d' % chrom
snp_filter = None
if snp_filter_frac < 1:
snp_filter = sp.random.random(len(in_h5f[chrom_str]['snps'])) < snp_filter_frac
g_dict = get_genotype_data(in_h5f, chrom, maf_thres, indiv_filter=indiv_filter,
snp_filter=snp_filter, randomize_sign=True, snps_signs=None, chrom_ok_snp_dict=chrom_ok_snp_dict)
norm_snps = g_dict['norm_snps']
sum_indiv_genotypes = sp.sum(g_dict['norm_snps'], 0)
sum_indiv_genotypes_all_chrom += sum_indiv_genotypes
print 'Calculating chromosome kinship'
K_unscaled = sp.array(sp.dot(norm_snps.T, norm_snps), dtype='float32')
assert sp.isclose(sp.sum(sp.diag(K_unscaled)) / (len(norm_snps) * num_indivs), 1.0), '..bug'
K_all_snps += K_unscaled
num_all_snps += len(norm_snps)
print 'SNP-cov normalisation'
sum_indiv_genotypes = sp.sum(norm_snps, 0)
sum_indiv_genotypes_all_chrom += sum_indiv_genotypes
mean_indiv_genotypes = sum_indiv_genotypes / len(norm_snps)
norm_snps = norm_snps - mean_indiv_genotypes
print 'Calculating SNP covariance unscaled'
snp_cov_unscaled = sp.array(sp.dot(norm_snps.T, norm_snps), dtype='float32')
# snp_cov_all_snps += snp_cov_unscaled
print 'Storing and updating things'
chromosome_dict[chrom_str] = {'K_unscaled':K_unscaled, 'num_snps':len(norm_snps),
'sum_indiv_genotypes':sum_indiv_genotypes,
'snp_cov_unscaled':snp_cov_unscaled,
'snps_signs':g_dict['snps_signs']}
if snp_filter_frac < 1:
chromosome_dict[chrom_str]['snp_filter'] = snp_filter
# snp_cov_all_snps = snp_cov_all_snps / float(num_all_snps)
# K_all_snps = K_all_snps / float(num_all_snps)
# print 'K_all_snps.shape: %s' % str(K_all_snps.shape)
# print 'snp_cov_all_snps.shape: %s' % str(snp_cov_all_snps.shape)
# print 'sp.diag(snp_cov_all_snps): %s' % str(sp.diag(snp_cov_all_snps))
# print 'sp.mean(sp.diag(snp_cov_all_snps)_: %s' % str(sp.mean(sp.diag(snp_cov_all_snps)))
# print 'Full kinship and snp-covariance calculation done using %d SNPs\n' % num_all_snps
mean_indiv_genotypes_all_chrom = sum_indiv_genotypes_all_chrom / num_all_snps
print 'Individual gentoype mean found:'
print mean_indiv_genotypes_all_chrom
print 'Calculating chromosome-wise SNP-covariance and kinship matrices'
for chrom in range(1, 23):
if chrom in not_done:
print 'Working on Chromosome %d' % chrom
chrom_str = 'chr%d' % chrom
snp_cov_leave_one_out = sp.zeros((num_indivs, num_indivs), dtype='float32')
K_leave_one_out = sp.zeros((num_indivs, num_indivs), dtype='float32')
num_snps_used = 0
sum_indiv_genotypes = sp.zeros(num_indivs, dtype='float32')
for chrom2 in range(1, 23):
chrom2_str = 'chr%d' % chrom2
if chrom2 != chrom:
sum_indiv_genotypes += chromosome_dict[chrom2_str]['sum_indiv_genotypes']
K_leave_one_out += chromosome_dict[chrom2_str]['K_unscaled']
num_snps_used += chromosome_dict[chrom2_str]['num_snps']
assert sp.isclose(sp.sum(sp.diag(K_leave_one_out)) / (num_snps_used * num_indivs), 1.0), '..bug'
mean_indiv_genotypes = sum_indiv_genotypes / num_snps_used
for chrom2 in range(1, 23):
chrom2_str = 'chr%d' % chrom2
if chrom2 != chrom:
print 'Loading SNPs'
snps_signs = chromosome_dict[chrom2_str]['snps_signs']
snp_filter = chromosome_dict[chrom2_str]['snp_filter']
g_dict = get_genotype_data(in_h5f, chrom2, maf_thres, indiv_filter=indiv_filter,
snp_filter=snp_filter, randomize_sign=True,
snps_signs=snps_signs, chrom_ok_snp_dict=chrom_ok_snp_dict)
norm_snps = g_dict['norm_snps']
print 'SNP-cov normalisation'
norm_snps = norm_snps - mean_indiv_genotypes
print 'Calculating SNP covariance unscaled'
snp_cov_unscaled = sp.dot(norm_snps.T, norm_snps)
snp_cov_leave_one_out += snp_cov_unscaled
snp_cov_leave_one_out = snp_cov_leave_one_out / num_snps_used
K_leave_one_out = K_leave_one_out / num_snps_used
assert (K_leave_one_out - sp.diag(K_leave_one_out)).max() < 0.1, '..bug'
try:
cholesky_decomp_inv_snp_cov = linalg.cholesky(linalg.pinv(sp.array(snp_cov_leave_one_out, dtype='float64')))
evals, evecs = linalg.eig(sp.array(K_leave_one_out, dtype='float64'))
except:
try:
cholesky_decomp_inv_snp_cov = linalg.cholesky(linalg.pinv(sp.array(snp_cov_leave_one_out, dtype='float32')))
evals, evecs = linalg.eig(sp.array(K_leave_one_out, dtype='float32'))
except:
print 'Failed when obtaining the Cholesky decomposotion or eigen decomposition'
print 'Moving on, trying again later.'
continue
sort_indices = sp.argsort(evals,)
ordered_evals = evals[sort_indices]
print ordered_evals[-10:] / sp.sum(ordered_evals)
ordered_evecs = evecs[:, sort_indices]
d = {}
d['evecs_leave_one_out'] = ordered_evecs
d['evals_leave_one_out'] = ordered_evals
d['cholesky_decomp_inv_snp_cov'] = cholesky_decomp_inv_snp_cov
d['K_leave_one_out'] = K_leave_one_out
d['K_unscaled'] = chromosome_dict[chrom_str]['K_unscaled']
d['num_snps'] = chromosome_dict[chrom_str]['num_snps']
d['snp_cov_leave_one_out'] = snp_cov_leave_one_out
ok_chromosome_dict[chrom_str] = d
not_done.remove(chrom)
# While loop ends here.
K_all_snps = K_all_snps / float(num_all_snps)
in_h5f.close()
ok_chromosome_dict['K_all_snps'] = K_all_snps
ok_chromosome_dict['num_all_snps'] = num_all_snps
assert sp.sum((ok_chromosome_dict['chr1']['K_leave_one_out'] - ok_chromosome_dict['chr2']['K_leave_one_out']) ** 2) != 0 , 'Kinships are probably too similar.'
print 'Calculating PCAs'
evals, evecs = linalg.eigh(sp.array(K_all_snps, dtype='float64')) # PCA via eigen decomp
evals[evals < 0] = 0
sort_indices = sp.argsort(evals,)[::-1]
ordered_evals = evals[sort_indices]
print ordered_evals[:10] / sp.sum(ordered_evals)
pcs = evecs[:, sort_indices]
tot = sum(evals)
var_exp = [(i / tot) * 100 for i in sorted(evals, reverse=True)]
print 'Total variance explained:', sp.sum(var_exp)
ok_chromosome_dict['pcs'] = pcs
ok_chromosome_dict['pcs_var_exp'] = var_exp
if figure_dir is not None:
plt.clf()
plt.plot(pcs[:, 0], pcs[:, 1], 'k.')
plt.title("Overall PCA")
plt.xlabel('PC1')
plt.xlabel('PC2')
plt.tight_layout()
plt.savefig(figure_dir + '/' + figure_fn, format='pdf')
plt.clf()
out_h5f = h5py.File(out_file)
hu.dict_to_hdf5(ok_chromosome_dict, out_h5f)
out_h5f.close()
return ok_chromosome_dict
