def test_n_splits(num_sub_phenos,
frac_shared_effects,
num_snps=100,
num_cases=5000,
num_conts=5000):
num_snps_shared = num_snps * frac_shared_effects
# ensure equal number of non-shared SNPs across sub-phenotypes
assert (num_snps - num_snps_shared) % num_sub_phenos == 0
num_snps_exclsv = (num_snps - num_snps_shared) / num_sub_phenos
h_sq = 0.034
ps = np.array([0.2] * num_snps)
cases = np.zeros((0, num_snps))
conts = np.zeros((0, num_snps))
for i in range(num_sub_phenos):
num_sub_cases = int(num_cases / num_sub_phenos)
num_sub_conts = num_sub_cases
sub_betas = np.array([ int(j < num_snps_shared or (
j>= num_snps_shared + i*(num_snps_exclsv) and \
j< num_snps_shared + (i+1)*(num_snps_exclsv))
) for j in range(num_snps)])
print(num_snps_shared)
print(num_snps_exclsv)
print(sub_betas)
# set variance explained over the subset
beta_val = np.sqrt(
h_sq / np.sum(np.multiply(2 * np.multiply(ps, 1 - ps), sub_betas)))
sub_betas = beta_val * sub_betas
print(i, num_sub_cases)
print(sub_betas)
print("subset variance explained:",
np.dot(np.square(sub_betas), 2 * np.multiply(ps, 1 - ps)))
# generate sub-cohort
prev = 0.01
thresh = norm.ppf(1 - prev, loc=0, scale=1)
sub_cases, sub_conts = generate_cohort(num_cases=num_sub_cases,
num_conts=num_sub_conts,
freqs=ps,
betas=sub_betas,
h_sq=h_sq,
thresh=thresh)
cases = np.concatenate((cases, sub_cases), axis=0)
conts = np.concatenate((conts, sub_conts), axis=0)
score = heterogeneity(cases, conts)
return score
def run_liability():
FILE_PATH = "subtype_frac_liab.p"
if os.path.exists(FILE_PATH):
h_sq_frac_range, results, hetsc_exp = pickle.load(open(
FILE_PATH, "rb"))
plot_results(h_sq_frac_range,
results,
hetsc_exp,
outname="subtype_scale_liability.eps",
xaxis_label='Fraction of subtype variance explained')
else:
num_cases = 30000
num_conts = 30000
num_snps = 100
fixed_ps = np.array([0.2] * num_snps)
# num_trials = 3
num_trials = 20
h_sq = 0.05
h_sq_frac_range = [0, 0.2, 0.4, 0.6, 0.8, 1.0]
sub_betas_list, _ = generate_snps_splits(num_sub_phenos=2,
frac_shared_effects=0,
num_snps=num_snps,
ps=fixed_ps,
h_sq=h_sq)
# subtype 1 properties
ps, betas = generate_snp_props(num_snps, fixed_ps, h_sq)
prev = 0.01
thresh = norm.ppf(1 - prev, loc=0, scale=1)
# expected score if all cases followed subtype 1
hetsc_exp = heterogeneity_expected_corr(ncases=num_cases,
nconts=num_cases,
effects=betas,
thresh=thresh,
freqs=fixed_ps,
heritability=h_sq,
verbose=False)
print(hetsc_exp)
results = []
for h_sq_frac in h_sq_frac_range:
scores = []
# generate subtype 2 properties
# _, betas_sub2 = generate_snp_props(num_snps, fixed_ps, h_sq*h_sq_frac)
sub_betas_list2, _ = generate_snps_splits(num_sub_phenos=2,
frac_shared_effects=0,
num_snps=num_snps,
ps=fixed_ps,
h_sq=h_sq * h_sq_frac)
for nt in range(num_trials):
print("h_sq_frac: %s, trial: %s" % (h_sq_frac, nt))
cases_sub1, conts = generate_cohort(
num_cases=int(num_cases / 2),
num_conts=num_conts,
freqs=fixed_ps, # freqs=ps,
betas=sub_betas_list[0], # betas=betas,
h_sq=h_sq,
thresh=thresh)
cases_sub2, _ = generate_cohort(
num_cases=int(num_cases / 2),
num_conts=num_conts,
freqs=fixed_ps, # freqs=ps,
betas=sub_betas_list2[1], # betas=betas_sub2,
h_sq=h_sq * h_sq_frac,
thresh=thresh)
cases = np.concatenate((cases_sub1, cases_sub2), axis=0)
score = heterogeneity(cases, conts)
scores.append(score)
results.append(scores)
pickle.dump((h_sq_frac_range, results, hetsc_exp),
open(FILE_PATH, "wb"))
def run_logistic():
FILE_PATH = "subtype_frac_logit.p"
if os.path.exists(FILE_PATH):
h_sq_frac_range, results, hetsc_exp = pickle.load(open(
FILE_PATH, "rb"))
plot_results(h_sq_frac_range,
results,
hetsc_exp,
outname="subtype_scale_logit.eps",
xaxis_label='Fraction of subtype odds ratio magnitude')
else:
num_cases = 30000
num_conts = 30000
num_snps = 100
fixed_ps = np.array([0.2] * num_snps)
# num_trials = 3
num_trials = 20
OR_val = 1.1
OR_val_frac_range = [0, 0.2, 0.4, 0.6, 0.8, 1.0]
prev = 0.01
# subtype 1 properties
# ORs = np.array([OR_val]*num_snps)
ORs = np.array([OR_val] * int(num_snps / 2) +
[1.0] * int(num_snps / 2))
# expected score if all cases followed subtype 1
hetsc_exp = heterogeneity_expected_corr_logit(ncases=num_cases,
nconts=num_cases,
ORs=ORs,
freqs=fixed_ps,
prev=prev,
verbose=False)
print(hetsc_exp)
results = []
for OR_val_frac in OR_val_frac_range:
scores = []
# generate subtype 2 properties
ORs_frac_val = 1.0 + (OR_val - 1) * OR_val_frac
# ORs_frac = np.array([ORs_frac_val]*num_snps)
ORs2 = np.array([1.0] * int(num_snps / 2) +
[ORs_frac_val] * int(num_snps / 2))
for nt in range(num_trials):
print("OR_val_frac: %s, trial: %s" % (OR_val_frac, nt))
cases_sub1, conts = generate_cohort_logistic(
num_cases=int(num_cases / 2),
num_conts=num_conts,
freqs=fixed_ps,
ORs=ORs,
prev=prev)
cases_sub2, _ = generate_cohort_logistic(num_cases=int(
num_cases / 2),
num_conts=num_conts,
freqs=fixed_ps,
ORs=ORs2,
prev=prev)
cases = np.concatenate((cases_sub1, cases_sub2), axis=0)
score = heterogeneity(cases, conts)
scores.append(score)
results.append(scores)
pickle.dump((OR_val_frac_range, results, hetsc_exp),
open(FILE_PATH, "wb"))
def run_heritability():
FILE_PATH = "simulate_results_varexp.p"
fig, ax = plt.subplots(1,1)
ax.spines['right'].set_visible(False)
ax.spines['top'].set_visible(False)
if os.path.exists(FILE_PATH):
pckl = pickle.load(open(FILE_PATH, "rb"))
# hetsc_means_hom = pckl["hetsc_means_hom"]
# hetsc_stds_hom = pckl["hetsc_stds_hom"]
# hetsc_means_het = pckl["hetsc_means_het"]
# hetsc_stds_het = pckl["hetsc_stds_het"]
# hetsc_means_cont = pckl["hetsc_means_cont"]
# hetsc_stds_cont = pckl["hetsc_stds_cont"]
hetsc_means_hom = [np.mean(x) for x in pckl["hetscs_homs"]]
hetsc_stds_hom = [np.std(x) for x in pckl["hetscs_homs"]]
hetsc_means_het = [np.mean(x) for x in pckl["hetscs_hets"]]
hetsc_stds_het = [np.std(x) for x in pckl["hetscs_hets"]]
hetsc_means_cont = [np.mean(x) for x in pckl["hetscs_conts"]]
hetsc_stds_cont = [np.std(x) for x in pckl["hetscs_conts"]]
hetsc_exps = pckl["hetsc_exps"]
h_sqs = pckl["h_sqs"]
plt.errorbar(h_sqs, hetsc_means_hom, yerr=hetsc_stds_hom, color='red', capsize=5)
plt.errorbar(h_sqs, hetsc_means_het, yerr=hetsc_stds_het, color='green', capsize=5)
plt.errorbar(h_sqs, hetsc_means_cont, yerr=hetsc_stds_cont, color='black', capsize=5)
plt.plot(h_sqs, hetsc_exps, color='blue')
plt.xlabel("Variance explained by modeled SNPs", fontsize=14)
plt.ylabel("Heterogeneity Score", fontsize=14)
# label CLiP score
xloc = 0.8 * (h_sqs[-1] - h_sqs[0])
yloc1 = np.interp(xloc, h_sqs, hetsc_exps)
yloc2 = np.interp(xloc, h_sqs, hetsc_means_het)
ax.annotate(s='',xy=(xloc, yloc1), xycoords='data',
xytext=(xloc, yloc2),textcoords='data',
arrowprops=dict(arrowstyle="<->", color='gray'))
ax.annotate(s='CLiP Score',xy=(xloc * 1.02, yloc1 + (yloc2-yloc1)*0.75),
xycoords='data',fontsize=14.0,textcoords='data',
ha='left', color='gray')
plt.xticks(fontsize=11)
plt.yticks(fontsize=11)
plt.savefig("simulate_basic_heritability.eps", format="eps", dpi=1000)
plt.show()
else:
# hetsc_means_hom = []
# hetsc_stds_hom = []
# hetsc_means_het = []
# hetsc_stds_het = []
# hetsc_means_cont = []
# hetsc_stds_cont = []
hetscs_homs = []
hetscs_hets = []
hetscs_conts = []
hetsc_exps = []
h_sqs = [0.001, 0.01, 0.02, 0.03, 0.05, 0.075, 0.1]
num_trials = 20
for h_sq in h_sqs:
num_cases = 30000
num_conts = 30000
num_snps = 100
fixed_ps = np.array([0.2]*num_snps)
hetscs_hom = []
hetscs_het = []
hetscs_cont = []
# expected score, homogeneous cases
ps, betas = generate_snp_props(num_snps, fixed_ps, h_sq)
prev = 0.01
thresh = norm.ppf(1-prev, loc=0, scale=1)
hetsc_exp = heterogeneity_expected_corr(ncases = num_cases,
nconts = num_conts,
effects = betas,
thresh = thresh,
freqs = ps,
heritability = h_sq, verbose=False)
hetsc_exps.append(hetsc_exp)
for i in range(num_trials):
print("h_sq: %s, trial: %s" % (h_sq, i))
# homogeneous cases, controls
cases,conts = generate_homhet_cohort(num_cases, num_conts, num_snps, ps, h_sq)
score = heterogeneity(cases,conts)
hetscs_hom.append(score)
# heterogeneous cases, controls
cases,conts = generate_homhet_cohort(num_cases, num_conts, num_snps, ps, h_sq, het=True)
score = heterogeneity(cases,conts)
hetscs_het.append(score)
# controls, controls
cases,conts = generate_controls(num_cases, num_conts, num_snps, ps, h_sq)
score = heterogeneity(cases,conts)
hetscs_cont.append(score)
# hetsc_means_hom.append(np.mean(hetscs_hom))
# hetsc_stds_hom.append(np.std(hetscs_hom))
# hetsc_means_het.append(np.mean(hetscs_het))
# hetsc_stds_het.append(np.std(hetscs_het))
# hetsc_means_cont.append(np.mean(hetscs_cont))
# hetsc_stds_cont.append(np.std(hetscs_cont))
hetscs_homs.append(hetscs_hom)
hetscs_hets.append(hetscs_het)
hetscs_conts.append(hetscs_cont)
pickle.dump({"hetscs_homs":hetscs_homs,
"hetscs_hets":hetscs_hets,
"hetscs_conts":hetscs_conts,
"hetsc_exps":hetsc_exps,
"h_sqs":h_sqs}, open(FILE_PATH, "wb"))
"""
def run_sample_size():
FILE_PATH = "simulate_results_samplesize.p"
fig, ax = plt.subplots(1,1)
ax.spines['right'].set_visible(False)
ax.spines['top'].set_visible(False)
if os.path.exists(FILE_PATH):
pckl = pickle.load(open(FILE_PATH, "rb"))
# hetsc_means_hom = pckl["hetsc_means_hom"]
# hetsc_stds_hom = pckl["hetsc_stds_hom"]
# hetsc_means_het = pckl["hetsc_means_het"]
# hetsc_stds_het = pckl["hetsc_stds_het"]
# hetsc_means_cont = pckl["hetsc_means_cont"]
# hetsc_stds_cont = pckl["hetsc_stds_cont"]
hetsc_means_hom = [np.mean(x) for x in pckl["hetscs_homs"]]
hetsc_stds_hom = [np.std(x) for x in pckl["hetscs_homs"]]
hetsc_means_het = [np.mean(x) for x in pckl["hetscs_hets"]]
hetsc_stds_het = [np.std(x) for x in pckl["hetscs_hets"]]
hetsc_means_cont = [np.mean(x) for x in pckl["hetscs_conts"]]
hetsc_stds_cont = [np.std(x) for x in pckl["hetscs_conts"]]
hetsc_exps = pckl["hetsc_exps"]
sample_size = pckl["sample_size"]
plt.errorbar(sample_size, hetsc_means_hom, yerr=hetsc_stds_hom, color='red', capsize=5)
plt.errorbar(sample_size, hetsc_means_het, yerr=hetsc_stds_het, color='green', capsize=5)
plt.errorbar(sample_size, hetsc_means_cont, yerr=hetsc_stds_cont, color='black', capsize=5)
# plot values
for i in range(len(sample_size)):
print("sample size: %s, score: %s, std dev: %s" % (sample_size[i], hetsc_means_hom[i], hetsc_stds_hom[i]))
plt.plot(sample_size, hetsc_exps, color='blue')
plt.xlabel("Number of simulated cases and controls", fontsize=14)
plt.ylabel("Heterogeneity Score", fontsize=14)
# label CLiP score
xloc = 0.8 * (sample_size[-1] - sample_size[0])
yloc1 = np.interp(xloc, sample_size, hetsc_exps)
yloc2 = np.interp(xloc, sample_size, hetsc_means_het)
ax.annotate(s='',xy=(xloc, yloc1), xycoords='data',
xytext=(xloc, yloc2),textcoords='data',
arrowprops=dict(arrowstyle="<->", color='gray'))
ax.annotate(s='CLiP Score',xy=(xloc * 1.02, yloc1 + (yloc2-yloc1)*0.75),
xycoords='data',fontsize=14.0,textcoords='data',
ha='left', color='gray')
plt.xticks(fontsize=11)
plt.yticks(fontsize=11)
plt.savefig("simulate_basic_num_inds.eps", format="eps", dpi=1000)
plt.show()
else:
# hetsc_means_hom = []
# hetsc_stds_hom = []
# hetsc_means_het = []
# hetsc_stds_het = []
# hetsc_means_cont = []
# hetsc_stds_cont = []
hetscs_homs = []
hetscs_hets = []
hetscs_conts = []
hetsc_exps = []
sample_sizes = [1000, 5000, 10000, 20000, 30000, 50000]
# sample_sizes = [5000, 10000]
# sample_sizes = [1000, 5000, 10000]
num_trials = 20
for sample_size in sample_sizes:
num_cases = sample_size
num_conts = sample_size
num_snps = 100
h_sq = 0.034
fixed_ps = np.array([0.2]*num_snps)
hetscs_hom = []
hetscs_het = []
hetscs_cont = []
# expected score, homogeneous cases
ps, betas = generate_snp_props(num_snps, fixed_ps, h_sq)
prev = 0.01
thresh = norm.ppf(1-prev, loc=0, scale=1)
hetsc_exp = heterogeneity_expected_corr(ncases = num_cases,
nconts = num_conts,
effects = betas,
thresh = thresh,
freqs = ps,
heritability = h_sq, verbose=False)
hetsc_exps.append(hetsc_exp)
for i in range(num_trials):
print("sample_size: %s, trial: %s" % (sample_size, i))
# homogeneous cases, controls
cases,conts = generate_homhet_cohort(num_cases, num_conts, num_snps, fixed_ps, h_sq)
score = heterogeneity(cases,conts)
hetscs_hom.append(score)
# heterogeneous cases, controls
cases,conts = generate_homhet_cohort(num_cases, num_conts, num_snps, fixed_ps, h_sq, het=True)
score = heterogeneity(cases,conts)
hetscs_het.append(score)
# controls, controls
cases,conts = generate_controls(num_cases, num_conts, num_snps, fixed_ps, h_sq)
score = heterogeneity(cases,conts)
hetscs_cont.append(score)
# hetsc_means_hom.append(np.mean(hetscs_hom))
# hetsc_stds_hom.append(np.std(hetscs_hom))
# hetsc_means_het.append(np.mean(hetscs_het))
# hetsc_stds_het.append(np.std(hetscs_het))
# hetsc_means_cont.append(np.mean(hetscs_cont))
# hetsc_stds_cont.append(np.std(hetscs_cont))
hetscs_homs.append(hetscs_hom)
hetscs_hets.append(hetscs_het)
hetscs_conts.append(hetscs_cont)
pickle.dump({"hetscs_homs":hetscs_homs,
"hetscs_hets":hetscs_hets,
"hetscs_conts":hetscs_conts,
"hetsc_exps":hetsc_exps,
"sample_size":sample_sizes}, open(FILE_PATH, "wb"))
"""
# frac_shared_effects_list = [0, .25, .5, .75, 1]
frac_shared_effects_list = [0, .5, 1]
split_results = {}
for i in num_subpheno_list:
split_results[i] = {}
for j in frac_shared_effects_list:
split_results[i][j] = []
for nsubph in num_subpheno_list:
for fsheff in frac_shared_effects_list:
for nt in range(num_trials):
print(nsubph, fsheff, nt)
h_sq = 0.05
sub_betas_list, ps = generate_snps_splits(
num_sub_phenos=nsubph,
frac_shared_effects=fsheff,
num_snps=num_snps,
ps=ps,
h_sq=h_sq)
cases, conts = generate_cohort_splits(
num_sub_phenos=nsubph,
sub_betas_list=sub_betas_list,
ps=ps,
num_cases=num_cases,
num_conts=num_cases,
h_sq=h_sq)
score = heterogeneity(cases, conts)
split_results[nsubph][fsheff].append(score)
pickle.dump(split_results, open("split_results.p", "wb"))
# pickle file for saving correlation tests
corr_matrices = []
# for display table and FDR calculation
table_vals = []
# for plotting score vs num_cases
plot_vals = []
score_std = 1
for fl in files:
cht_name, FILE_PATH, SAMPLE_PATH = fl
# try:
cases, conts, cases_abr, conts_abr, snpsdel, ors, frqs = extract(
SNP_PATH, FILE_PATH, SAMPLE_PATH)
hetsc = heterogeneity(cases_abr, conts_abr)
# convert odds ratios to liability threshold ratios
betas = convertORs(ors, prev)
h_sq = np.sum(
np.multiply(np.square(betas), 2 * np.multiply(frqs, 1 - frqs)))
# print("h_sq:", h_sq)
expected_score = heterogeneity_expected_corr(ncases=cases.shape[0],
nconts=conts.shape[0],
effects=betas,
thresh=thresh,
freqs=frqs,
heritability=h_sq,
verbose=False)
if snpsdel == 0: # store for combined set
super_betas = betas
def run_heritability():
# FILE_PATH = "population_stratif_out.p"
# pop_allele_diffs = [0.0, 0.05, 0.1, 0.15, 0.2]
fsts = np.array([0.001, 0.005, 0.01, 0.05, 0.1])
pop_allele_diffs = 0.5*np.sqrt(fsts)
# for allele_diff in pop_allele_diffs:
# FILE_PATH = "population_stratif_out_%s.p" % (allele_diff)
for fst, allele_diff in zip(fsts, pop_allele_diffs):
FILE_PATH = "population_stratif_out_%s.p" % (fst)
if not os.path.exists(FILE_PATH):
num_snps = 100
fixed_ps_val = 0.5
fixed_ps = np.array([fixed_ps_val]*num_snps)
# generate stratified populations
ps_stratif = [[fixed_ps_val - allele_diff]*int(num_snps/2) + [fixed_ps_val + allele_diff]*int(num_snps/2),
[fixed_ps_val + allele_diff]*int(num_snps/2) + [fixed_ps_val - allele_diff]*int(num_snps/2)]
hetscs_homs = []
hetscs_hets = []
hetscs_conts = []
hetsc_exps = []
mean_allele_diffs = [] # allele difference between homogeneous cases and controls
h_sqs = [0.001, 0.01, 0.02, 0.03, 0.05, 0.075, 0.1]
# h_sqs = [0.001, 0.025, 0.05, 0.075, 0.1]
num_trials = 30
for h_sq in h_sqs:
num_cases = 30000
num_conts = 30000
hetscs_hom = []
hetscs_het = []
hetscs_cont = []
mean_allele_diff = []
# expected score, homogeneous cases
ps, betas = generate_snp_props(num_snps, fixed_ps, h_sq)
prev = 0.01
thresh = norm.ppf(1-prev, loc=0, scale=1)
hetsc_exp = heterogeneity_expected_corr(ncases = num_cases,
nconts = num_conts,
effects = betas,
thresh = thresh,
freqs = ps,
heritability = h_sq, verbose=False)
hetsc_exps.append(hetsc_exp)
for i in range(num_trials):
print("allele_diff: %s, h_sq: %s, trial: %s" % (allele_diff, h_sq, i))
# homogeneous cases, controls
cases,conts = generate_homhet_cohort(num_cases, num_conts, num_snps, ps, ps_stratif, h_sq)
score = heterogeneity(cases,conts)
hetscs_hom.append(score)
# heterogeneous cases, controls
cases,conts = generate_homhet_cohort(num_cases, num_conts, num_snps, ps, ps_stratif, h_sq, het=True)
score = heterogeneity(cases,conts)
hetscs_het.append(score)
mean_allele_diff.append(np.mean(cases) - np.mean(conts))
# controls, controls
# cases,conts = generate_controls(num_cases, num_conts, num_snps, ps, h_sq)
cases, conts = generate_controls(num_cases, num_conts, num_snps, ps_stratif)
score = heterogeneity(cases,conts)
hetscs_cont.append(score)
hetscs_homs.append(hetscs_hom)
hetscs_hets.append(hetscs_het)
hetscs_conts.append(hetscs_cont)
mean_allele_diffs.append(mean_allele_diff)
pickle.dump({"hetscs_homs":hetscs_homs,
"hetscs_hets":hetscs_hets,
"hetscs_conts":hetscs_conts,
"hetsc_exps":hetsc_exps,
"mean_allele_diffs":mean_allele_diffs,
"h_sqs":h_sqs}, open(FILE_PATH, "wb"))
plot_results(fsts)
