def get_data_with_bucket_score(yzer, dirpath):
filename = yzer.get_filename(
dirpath, 'refseq_by_transcript_and_bucket_with_lfc.txt')
data = yzer.import_file(filename)
data = data.fillna(0)
run_ids = set_up_sequencing_run_ids()
total_tags = total_tags_per_run()
# For each sequencing run group, fill in the bucket score val
for replicate_id in ('', 1, 2, 3, 4):
for run_type, id_set in run_ids.iteritems():
rep_str = get_rep_string(replicate_id)
curr_ids = id_set[replicate_id or 0]
# First subset rows by sequencing run_ids of interest
subset = data[data['sequencing_run_id'].isin(curr_ids)]
# Group by transcript, and assign bucket score to each row
grouped = subset.groupby('glass_transcript_id')
bucket_scores = grouped.apply(bucket_score)
gene_start_sums = grouped.apply(gene_start_tags)
gene_body_sums = grouped.apply(gene_body_tags)
# Fill in bucket score for each original row.
data['{0}_{1}bucket_score'.format(
run_type,
rep_str)] = bucket_scores[data['glass_transcript_id']].values
data['{0}_{1}gene_start_tags'.format(
run_type,
rep_str)] = gene_start_sums[data['glass_transcript_id']].values
data['{0}_{1}gene_body_tags'.format(
run_type,
rep_str)] = gene_body_sums[data['glass_transcript_id']].values
# Fill NAs created by transcripts missing from certain sequencing runs.
data['{0}_{1}bucket_score'.format(run_type, rep_str)] = \
data['{0}_{1}bucket_score'.format(run_type, rep_str)].fillna(1) # Use 1 to show parity.
data['{0}_{1}gene_start_tags'.format(run_type, rep_str)] = \
data['{0}_{1}gene_start_tags'.format(run_type, rep_str)].fillna(0)
data['{0}_{1}gene_body_tags'.format(run_type, rep_str)] = \
data['{0}_{1}gene_body_tags'.format(run_type, rep_str)].fillna(0)
# Now calculate gene body log fold change
kla_norm = total_tags['dmso'][replicate_id
or 0] / total_tags['kla'][replicate_id
or 0]
dex_kla_norm = total_tags['kla'][
replicate_id or 0] / total_tags['kla_dex'][replicate_id or 0]
data['kla_{0}gene_body_lfc'.format(rep_str)] = data.apply(
lambda x: gene_body_lfc(x, kla_norm, rep_str, 'kla', 'dmso'),
axis=1)
data['dex_over_kla_{0}gene_body_lfc'.format(rep_str)] = data.apply(
lambda x: gene_body_lfc(x, dex_kla_norm, rep_str, 'kla_dex', 'kla'
),
axis=1)
return data
def draw_boxplot(data, label, dirpath):
curr_dirpath = grapher.get_filename(dirpath, 'boxplots_{0}'.format(label))
if not os.path.exists(curr_dirpath): os.mkdir(curr_dirpath)
# Group and take mean. We can use mean because we only care
# about the values that are the same for all rows for a given transcript.
data = data.groupby('glass_transcript_id', as_index=False).mean()
states = (
('KLA', 'kla_{0}state'),
('KLA+Dex', 'kla_dex_{0}state'),
('KLA+Dex over KLA', 'dex_over_kla_{0}state'),
)
for desc, state in states:
for replicate_id in ('', 1, 2, 3, 4):
rep_str = get_rep_string(replicate_id)
state_str = state.format(rep_str)
datasets = [
('No change', data[data[state_str] == 0]),
('Up >= 2x', data[data[state_str] == 1]),
('Down >= 2x', data[data[state_str] == -1]),
]
bar_names, datasets = zip(*datasets)
pausing_ratios = [
d['kla_dex_{0}bucket_score'.format(rep_str)] /
d['kla_{0}bucket_score'.format(rep_str)] for d in datasets
]
ax = grapher.boxplot(
pausing_ratios,
bar_names,
title='Pausing Ratio Delta in {0}, {1}'.format(
desc, (replicate_id and 'Group {0}'.format(replicate_id)
or 'Overall')),
xlabel='State in {0} {1}'.format(desc, replicate_id),
ylabel=
'Pausing ratio delta: (KLA+Dex pausing ratio)/(KLA pausing ratio)',
show_outliers=False,
show_plot=False)
pyplot.text(.05,
.9,
'Total transcripts: {0}'.format(len(data)),
transform=ax.transAxes)
grapher.save_plot(
grapher.get_filename(curr_dirpath,
'boxplot_{0}.png'.format(state_str)))
def get_tag_proportions(data, label):
total_trans = len(set(data['glass_transcript_id']))
print '\n\n\nTotal transcripts in {0}:\t{1}'.format(label, total_trans)
states = (
('KLA', 'kla_{0}state'),
('KLA+Dex', 'kla_dex_{0}state'),
('KLA+Dex over KLA', 'dex_over_kla_{0}state'),
)
for desc, state in states:
for replicate_id in ('', 1, 2, 3, 4):
rep_str = get_rep_string(replicate_id)
state_str = state.format(rep_str)
if desc == 'KLA':
datasets = [
('Up > 2x in KLA, Down > 1.5x from that in Dex {0}'.format(
replicate_id),
data[(data['kla_{0}state'.format(rep_str)] == 1)
& (data['dex_over_kla_{0}state'.format(rep_str)] ==
-1)]),
]
else:
datasets = []
datasets += [
('No change in {0} {1}'.format(desc, replicate_id),
data[data[state_str] == 0]),
('Up in {0} {1}'.format(desc, replicate_id),
data[data[state_str] == 1]),
('Down in {0} {1}'.format(desc, replicate_id),
data[data[state_str] == -1]),
]
for name, dataset in datasets:
dataset_trans = len(set(dataset['glass_transcript_id']))
print 'Total transcripts in {0}:\t{1}'.format(
name, dataset_trans)
print 'Percent of transcripts in {0}:\t{1}'.format(
name, dataset_trans / total_trans)
def draw_elongation_profile(data,
grapher,
dirpath,
show_moving_average=True,
show_count=False):
run_ids = set_up_sequencing_run_ids()
total_tags = total_tags_per_run()
states = (
('Special', 'group_{0}'),
('KLA', 'kla_{0}lfc'),
('KLA+Dex', 'kla_dex_{0}lfc'),
('KLA+Dex over KLA', 'dex_over_kla_{0}lfc'),
)
for desc, state in states:
for replicate_id in ('', 1, 2, 3, 4):
rep_str = get_rep_string(replicate_id)
state_str = state.format(rep_str)
# Include all transcripts at once, but only do it once.
if desc == 'Special':
datasets = [
('All RefSeq', data),
('Up > 2x in KLA, Down > 1.5x from that in Dex',
data[(data['kla_{0}lfc'.format(rep_str)] >= 1)
& (data['dex_over_kla_{0}lfc'.format(rep_str)] <=
-.58)]),
]
else:
datasets = [
('No change in {0}'.format(desc),
data[data[state_str].abs() < 1]),
('Up in {0}'.format(desc), data[data[state_str] >= 1]),
('Down in {0}'.format(desc), data[data[state_str] <= -1]),
]
for label, dataset in datasets:
slug_label = label.lower().replace(' ', '_')
group_by_cols = ['basepair', 'sequencing_run_id']
data_grouped = dataset.groupby(group_by_cols,
as_index=False).sum()
groups = [
data_grouped[data_grouped['sequencing_run_id'].isin(
run_ids['dmso'][replicate_id or 0])],
data_grouped[data_grouped['sequencing_run_id'].isin(
run_ids['kla'][replicate_id or 0])],
data_grouped[data_grouped['sequencing_run_id'].isin(
run_ids['kla_dex'][replicate_id or 0])]
]
# Combine for sequencing runs now
for i, group in enumerate(groups):
groups[i] = group.groupby(['basepair'],
as_index=False).sum()
totals = zip(*total_tags.values())[replicate_id or 0]
tag_scalars = grapher.get_tag_scalars(totals)
ax = grapher.plot_tags_per_basepair(
groups,
labels=['DMSO', 'KLA', 'KLA+Dex'],
title='Tag localization for RefSeq: {0}, {1}'.format(
label,
replicate_id and 'Group {0}'.format(replicate_id)
or 'overall'),
tag_scalars=tag_scalars,
show_moving_average=show_moving_average,
show_count=show_count)
grapher.save_plot(
grapher.get_filename(
dirpath, '{0}_refseq_by_run_type_{1}.png'.format(
slug_label,
replicate_id and 'group_{0}'.format(replicate_id)
or 'all')))
if True:
data = grapher.import_file(
grapher.get_filename(dirpath,
'transcripts_per_run_per_bucket.txt'))
# For the sake of graphing, imitate basepair
data['basepair'] = (data['bucket'] - 1) * 50
# Merge on transcripts to get DMSO tags.
transcripts = grapher.import_file(
grapher.get_filename(dirpath, 'transcript_vectors.txt'))
transcripts = transcripts[transcripts['has_refseq'] == 1]
transcripts['glass_transcript_id'] = transcripts['id']
totals = total_tags_per_run()
for replicate_id in ('', 1, 2, 3, 4):
rep_str = get_rep_string(replicate_id)
transcripts['dmso_{0}rpkm'.format(rep_str)] = transcripts['dmso_{0}tag_count'.format(rep_str)]*(10**3*10**6)\
/transcripts['length']/totals['dmso'][replicate_id or 0]
transcripts = transcripts.filter(
regex=r'glass_transcript_id|has_refseq|.*lfc|.*rpkm')
data = data.merge(transcripts, how='left', on='glass_transcript_id')
data = data.fillna(0)
data = data[data['has_refseq'] == 1]
# Create filtered groups.
datasets = [
('all_refseq', data),
('low_basal', data[data['dmso_rpkm'] < 1]),
('low_basal_1', data[data['dmso_1_rpkm'] < 1]),