ds.counts.log(inplace=True)
# Only select cells without virus
ds.query_samples_by_metadata('virus_reads_per_million < 0.1', inplace=True)
# Check table with number of cells
table = (ds.samplesheet.groupby(
['time', 'MOI']).count().iloc[:,
0].unstack().fillna(0).astype(int).loc[[
'4', '12', '24', '48'
]])
print('Selecting only early 2 time points')
# The rest has too few uninfected cells
ds.query_samples_by_metadata('time in ["4", "12"]', inplace=True)
dsm = ds.split('MOI')
ks = dsm['0'].compare(dsm['1'])['P-value']
# Get the top hits for GO analysis
hits = ds.featuresheet.loc[ks.nsmallest(100).index, 'GeneName'].values
with open('../tables/bystander_top_100.tsv', 'w') as f:
f.write('\n'.join(hits))
sys.exit()
# Bonferroni correction
ks = np.minimum(1, ks * len(ks))
# Print cumulative histogram of P-values
x = ks.sort_values().values
y = 1.0 - np.linspace(0, 1, len(x))
# Find the two clouds
dso = Dataset(
counts_table='dengue',
samplesheet='dengue',
featuresheet='humanGC38',
)
dso.query_samples_by_name(ds.samplenames, inplace=True)
dso.counts.normalize(inplace=True)
dso.feature_selection.unique(inplace=True)
dso.reindex(axis='features', column='GeneName', inplace=True, drop=False)
dso.samplesheet['virus_reads_per_million'] = ds.samplesheet['virus_reads_per_million']
ind_2424 = afs.sel(sample=samplenames, position=2424).fillna(-1).data >= 0.1
ind_not2424 = afs.sel(sample=samplenames, position=2424).fillna(-1).data < 0.1
dso.samplesheet['is_2424'] = ind_2424
dsp = dso.split('is_2424')
fig, ax = plt.subplots(figsize=(3.8, 3.2))
colors = {True: 'steelblue', False: 'darkred'}
for key, dsi in dsp.items():
color = colors[key]
if key is True:
label = 'M2'
else:
label = 'M1'
y = np.log10(0.1 + dsi.counts.loc['DDIT3'].values)
x = np.log10(0.1 + np.random.normal(0, 0.1, size=len(y)) + dsi.samplesheet['virus_reads_per_million'].values)
ax.scatter(x, y, s=10, color=color, alpha=0.15, label=label)
ax.grid(True)
ax.legend(loc='upper left', title='Mutant:')
ax.set_xticks([-1, 1, 3, 5])
ax.set_yticks([-1, 1, 3, 5])
ds.rename(axis='features', column='GeneName', inplace=True)
ds.feature_selection.unique(inplace=True)
ds.counts.normalize(inplace=True)
## Restrict to high variance SNVs
#dsv = ds.copy()
#ind = ds.counts.values.var(axis=1).argsort()[-200:]
#dsv.counts = dsv.counts.iloc[ind]
# Find upregulated genes
clusters = np.unique(ds.samplesheet['clusterN_SNV'])
genes = {}
for ic in clusters:
ds.samplesheet['clusterN_SNV_{:}'.format(
ic)] = ds.samplesheet['clusterN_SNV'] == ic
dss = ds.split('clusterN_SNV_{:}'.format(ic))
comp = dss[True].compare(dss[False])
# FIXME: maybe look symmetrically for up- and downregulated
comp['diff'] = dss[True].counts.mean(axis=1) - dss[False].counts.mean(
axis=1)
genesi = comp.loc[comp['diff'] > 0,
'P-value'].nsmallest(n=5).index.values
genes[ic] = genesi
genes_all = np.unique(np.concatenate(list(genes.values())))
with open('../data/genes_diff_expressed_clustersSNV.tsv', 'wt') as f:
f.write('\t'.join(genes_all))
dsv = ds.query_features_by_name(genes_all)
sharex=True,
sharey=True,
figsize=(8, 4))
dsr.plot.scatter_reduced_samples(vs,
color_by='dbscan',
ax=axs[0],
zorder=10)
dsr.plot.scatter_reduced_samples(vs,
color_by='kmeans',
ax=axs[1],
zorder=10)
axs[0].set_title('DBSCAN')
axs[1].set_title('K-means, 7 clusters')
plt.tight_layout()
ds.samplesheet['cluster'] = dsr.samplesheet['kmeans']
ds_dict = ds.split(phenotypes=['cluster'])
genes_by_cluster = {}
for key, dsi in ds_dict.items():
dso = ds.query_samples_by_metadata('cluster!=@key',
local_dict=locals())
genes_by_cluster[key] = dsi.compare(dso)['P-value'].nsmallest(10).index
assert (genes_by_cluster[1][:3].tolist() == [
'ENSG00000138085', 'ENSG00000184076', 'ENSG00000116459'
])
plt.show()
