def test_lambda_gc(self):
N = 5000000
ht = hl.utils.range_table(N).annotate(x=hl.scan.count() / N,
x2=(hl.scan.count() / N)**1.5)
lgc = hl.lambda_gc(ht.x)
lgc2 = hl.lambda_gc(ht.x2)
self.assertAlmostEqual(lgc, 1,
places=1) # approximate, 1 place is safe
self.assertAlmostEqual(lgc2, 1.89,
places=1) # approximate, 1 place is safe
def qqplot(pvals, title: str = None, figsize: tuple = (10, 10)):
source = pvals._indices.source
if isinstance(source, Table):
ht = source.select(p_value=pvals)
else:
ht = source.select_rows(p_value=pvals).rows()
ht = ht.key_by().select('p_value').key_by('p_value').persist()
lambda_gc = hl.lambda_gc(ht['p_value'])
n = ht.count()
ht = ht.annotate(
observed_p=-hl.log10(ht['p_value']),
expected_p=-hl.log10((hl.scan.count() + 1) / n),
p_val=ht['p_value']
).persist()
p_val_pd = ht.to_pandas()
p_val_pd['observed_p'].values[p_val_pd['observed_p'] > 10] = 10
mini = min(p_val_pd['expected_p'].max(), p_val_pd['observed_p'].max())
maxi = max(p_val_pd['expected_p'].max(), p_val_pd['observed_p'].max())
title = f'{title}' if title else 'QQ Plot'
fig = plt.figure(figsize=figsize)
plt.scatter(p_val_pd['expected_p'], p_val_pd['observed_p'], c='black', s=0.5)
plt.plot((0, mini), (0, mini), 'red')
plt.xlim([0, maxi + 0.5])
plt.ylim([0, maxi + 0.5])
plt.title(title, fontsize=20)
plt.ylabel('Observed -log10(' + r'$p$' + ')', fontsize=15)
plt.xlabel('Expected -log10(' + r'$p$' + ')', fontsize=15)
plt.close()
return fig, round(lambda_gc, 3)