def get_broad_sense_heritability(self, pids=None):
"""
Estimates the broad sense heritability from replicates.
"""
import linear_models as lm
if not pids:
pids = sorted(self.phen_dict.keys())
bs_herits = []
bs_avg_herits = []
bs_pids = []
bs_herit_pvals = []
for pid in pids:
ets = map(int, self.get_ecotypes(pid))
num_vals = len(ets)
num_ets = len(set(ets))
if num_vals == num_ets:
print "Can't estimate the broad sense heritability when replicates are missing."
continue
else:
avg_repl_num = float(num_vals) / num_ets
print 'Average number of replicates:', avg_repl_num
values = self.get_values(pid)
mod = lm.LinearModel(values)
res = mod.anova_f_test([sp.array(ets)])
bs_herit = res['var_perc'][0]
bs_herit_pval = res['ps'][0]
bs_herit_pvals.append(bs_herit_pval)
print 'Heritability:', bs_herit
print 'Heritability (different from 0) p-value :', bs_herit_pval
bs_avg_herit = 1.0 - (1.0 - bs_herit) / avg_repl_num
bs_avg_herits.append(bs_avg_herit)
print 'Estimated mean value heritability:', bs_avg_herit
bs_herits.append(bs_herit)
bs_pids.append(pid)
return {'bs_herits':bs_herits, 'bs_pids':bs_pids, 'bs_avg_herits':bs_avg_herits, 'bs_herit_pvals':bs_herit_pvals}
def get_broad_sense_heritability(self):
"""
Estimates the broad sense heritability from replicates.
"""
import linear_models as lm
ets = map(int, self.ecotypes)
num_vals = len(ets)
num_ets = len(set(ets))
if num_vals == num_ets:
log.warning(
"Can't estimate the broad sense heritability when replicates are missing."
)
else:
avg_repl_num = float(num_vals) / num_ets
log.info('Average number of replicates:%s' % avg_repl_num)
values = self.values
mod = lm.LinearModel(values)
res = mod.anova_f_test([sp.array(ets)])
bs_herit = res['var_perc'][0]
bs_herit_pval = res['ps'][0]
log.info('Heritability:%s' % bs_herit)
log.info('Heritability (different from 0) p-value :%s' % bs_herit_pval)
bs_avg_herit = 1.0 - (1.0 - bs_herit) / avg_repl_num
log.info('Estimated mean value heritability:%s' % bs_avg_herit)
return {
'bs_herit': bs_herit,
'bs_pid': bs_pid,
'bs_avg_herit': bs_avg_herit,
'bs_herit_pval': bs_herit_pval
}
def get_blup(self, pid, K):
"""
Returns the REML estimate for the BLUP and the pseudo-heritability.
"""
from scipy import stats
import linear_models as lm
phen_vals = self.get_values(pid)
lmm = lm.LinearMixedModel(phen_vals)
if len(phen_vals) == len(set(phen_vals)):
lmm.add_random_effect(K)
else:
Z = self.get_incidence_matrix(pid)
lmm.add_random_effect(Z * K * Z.T)
r1 = lmm.get_REML()
ll1 = r1['max_ll']
rlm = lm.LinearModel(phen_vals)
ll0 = rlm.get_ll()
lrt_stat = 2 * (ll1 - ll0)
pval = stats.chi2.sf(lrt_stat, 1)
#Now the BLUP.
y_mean = sp.mean(lmm.Y)
Y = lmm.Y - y_mean
p_herit = r1['pseudo_heritability']
delta = (1 - p_herit) / p_herit
# if K_inverse == None:
# K_inverse = K.I
# M = (sp.eye(K.shape[0]) + delta * K_inverse)
# u_blup = M.I * Y
M = (K + delta * sp.eye(K.shape[0]))
u_mean_pred = K * (M.I * Y)
blup_residuals = Y - u_mean_pred
return {'pseudo_heritability':r1['pseudo_heritability'], 'pval':pval, 'u_blup':u_mean_pred, 'blup_residuals':blup_residuals}
def get_pseudo_heritability(self, K):
"""
Returns the REML estimate of the heritability.
methods: 'avg' (averages), 'repl' (replicates)
"""
from scipy import stats
import linear_models as lm
lmm = lm.LinearMixedModel(self.phen_vals)
if len(self.values) == len(set(self.values)):
lmm.add_random_effect(K)
else:
Z = self.get_incidence_matrix()
lmm.add_random_effect(Z * K * Z.T)
r1 = lmm.get_REML()
ll1 = r1['max_ll']
rlm = lm.LinearModel(self.values)
ll0 = rlm.get_ll()
lrt_stat = 2 * (ll1 - ll0)
pval = stats.chi2.sf(lrt_stat, 1)
return {'pseudo_heritability': r1['pseudo_heritability'], 'pval': pval}