def main():
(_, chrom, antisense) = \
tuple(sys.argv)
antisense = antisense.lower() == 'true'
chrom = int(chrom)
print_fl("Running nucleosome calling on chromosome %d, antisense: %s" %
(chrom, str(antisense)))
name = task_name(antisense)
timer = Timer()
p123_orfs = paper_orfs
save_chrom_dir = sense_nuc_chrom_dir
cc_dir = cc_sense_chrom_dir
if antisense:
p123_orfs = antisense_orfs
save_chrom_dir = anti_nuc_chrom_dir
cc_dir = cc_antisense_chrom_dir
call_nucleosomes_p123_chrom(p123_orfs, chrom, antisense,
cc_dir, save_chrom_dir, timer)
child_done(name, WATCH_TMP_DIR, chrom)
def child_done(name, parent_watch_dir, child_name):
# write child done file
watch_dir = parent_watch_dir + '/' + name
write_path = ('%s/child_%s.watch' % (watch_dir, str(child_name)))
print_fl(write_path)
with open(write_path, 'wb') as f:
f.write('Done')
def wait_for_tasks(self):
"""
Wait for the number of child processes to finish by monitoring watch
directory
"""
while True:
# count # of child processes done
files = os.listdir(self.watch_dir)
num_done = sum([f.endswith('.watch') for f in files])
print_fl("%d/%d finished. " % (num_done, self.num_wait), end='')
if self.timer is not None:
print_fl(" %s " % self.timer.get_time(), end='')
# number of children done
if num_done == self.num_wait:
print_fl("All children done.")
break
# sleep until all children are done
else:
print_fl("Sleeping %ds" %
(self.sleep_time))
time.sleep(self.sleep_time)
# clean up watch directory
print_fl("Cleaning up watch directory.")
self.cleanup()
def select_data(self, head=500):
act_prm = self.store.promoter_sm_occupancy_delta
act_dorg = self.store.gene_body_disorganization_delta
mean_prm = act_prm.mean(axis=1).sort_values(ascending=False)
mean_dorg = act_dorg.mean(axis=1).sort_values(ascending=False)
prm_orfs = mean_prm.head(head).index
dorg_orfs = mean_dorg.head(head).index
self.data = self.store.chromatin_data.loc[set(prm_orfs).union(
set(dorg_orfs))]
def _inverse_quantile(array, val):
"""What is the quantile of the value in the array. (CDF)"""
return np.mean(array <= val)
prom_quantile = _inverse_quantile(mean_prm.values,
mean_prm.values[head])
dorg_quantile = _inverse_quantile(mean_dorg.values,
mean_dorg.values[head])
print_fl("Promoter ORFs: %d (%.1f%%)\n"
"Disorganization ORFs: %d (%.1f%%)" % \
(len(prm_orfs), prom_quantile*100.,
len(dorg_orfs), dorg_quantile*100.
))
print_fl("%d genes total" % len(self.data))
def compute_boundaries_chrom(orfs, pileup, chrom, save_dir, find_antisense,
log=False, timer=None):
chrom_genes = orfs[orfs.chr == chrom]
transcript_boundaries = chrom_genes[[]].copy()
transcript_boundaries['start'] = None
transcript_boundaries['stop'] = None
search_window = 2000
search_2 = search_window/2
if log and timer is not None:
print_fl("Chromosome %d - %s. %d genes" %
(chrom, timer.get_time(), len(chrom_genes)))
chrom_rna_seq = filter_rna_seq_pileup(pileup, chrom == chrom)
i = 0
for orf_name, gene in chrom_genes.iterrows():
if log and timer is not None and i % 100 == 0:
print_fl("%d/%d - %s" % (i, len(chrom_genes),
timer.get_time()))
i += 1
span = gene.transcript_start-search_2, gene.transcript_stop+search_2
gene_pileup = filter_rna_seq_pileup(chrom_rna_seq,
span[0], span[1], gene.chr)
try:
start, stop = find_transcript_boundaries(gene_pileup, span, gene,
find_antisense=find_antisense)
except ValueError:
# skip if issues finding boundaries
continue
transcript_boundaries.loc[orf_name, 'start'] = start
transcript_boundaries.loc[orf_name, 'stop'] = stop
TSS, TES = start, stop
if ((gene.strand == '-' and not find_antisense) or
(gene.strand == '+' and find_antisense)):
TSS, TES = stop, start
strand = gene.strand
if find_antisense: strand = flip_strand(gene.strand)
transcript_boundaries.loc[orf_name, 'TSS'] = TSS
transcript_boundaries.loc[orf_name, 'TES'] = TES
transcript_boundaries.loc[orf_name, 'strand'] = strand
transcript_boundaries.loc[orf_name, 'chr'] = gene.chr
transcript_boundaries = transcript_boundaries.dropna()
transcript_boundaries.to_csv(boundary_file_name(save_dir, chrom,
find_antisense))
return transcript_boundaries
def fit(self):
np.random.seed(123)
print_fl("Fitting positional distribution...")
self.fit_position()
print_fl("\nFitting length distribution...")
self.fit_length()
def plot_example_cross(plotter, save_dir):
from src.chromatin import filter_mnase
from src.plot_utils import apply_global_settings
from config import cross_corr_sense_path
from src.utils import get_orf
orf_cc = pd.read_hdf(cross_corr_sense_path, 'cross_correlation')
orfs = plotter.orfs
gene = get_orf('APJ1', orfs)
span = (gene.TSS - 500, gene.TSS + 500)
plotter.set_span_chrom(span, gene.chr)
cc_nuc = orf_cc.loc['nucleosomal'].loc[gene.name].loc[0.0]
cc_small = orf_cc.loc['small'].loc[gene.name].loc[0.0]
data = filter_mnase(plotter.all_mnase_data,
span[0],
span[1],
chrom=gene.chr,
time=0)
fig, (ax, leg_ax) = plt.subplots(2, 1, figsize=(5, 6))
fig.tight_layout(rect=[0.1, 0.1, 0.92, 0.945])
plt.subplots_adjust(hspace=0.0, wspace=0.5)
plotter.plot_typhoon_time(ax, data, 0, scale_z=True)
ax.set_xlim(*span)
ax.set_xticks(np.arange(span[0], span[1], 500))
ax.set_xticks(np.arange(span[0], span[1], 100), minor=True)
ax.set_xlabel("Position (bp)", fontsize=16)
ax.set_ylabel("Fragment length (bp)", fontsize=16, labelpad=10)
ax.set_ylim(-100, 250)
draw_legend(leg_ax, span, 500)
cc_ax = ax.twinx()
cc_ax.set_ylabel("Cross correlation $\\times$0.1",
rotation=270,
fontsize=16,
labelpad=10,
va='bottom')
scale_cc = 1
y_origin = 0
x = cc_nuc.index + gene.TSS
y = cc_nuc.values * scale_cc + y_origin
cc_ax.fill_between(x, y, y_origin, color='#28a098')
y = -cc_small.values * scale_cc + y_origin
cc_ax.fill_between(x, y_origin, y, color='#f28500')
cc_ax.set_ylim(-0.1, 0.4)
cc_ax.set_yticklabels(np.arange(-1, 5))
write_path = '%s/%s.pdf' % (save_dir, 'example_cross_correlation')
plt.savefig(write_path, transparent=True)
print_fl("Wrote %s" % write_path)
def danpos():
from src.dpos_bed import create_bed_for_dpos
import os
from src.utils import run_cmd
working_dir = os.getcwd()
danpos_output = '%s/danpos/' % (OUTPUT_DIR)
mkdirs_safe([danpos_output])
danpos_path = "%s/danpos-2.2.2/danpos.py" % working_dir
# create DANPOS Bed file
mnase = pd.read_hdf(mnase_seq_path, 'mnase_data')
mnase = mnase[mnase.time == 0]
save_file = 'mnase_0.bed'
save_path = '%s/%s' % (danpos_output, save_file)
create_bed_for_dpos(mnase, save_path)
print_fl("Wrote %s" % save_path)
bash_command = "scripts/6_reviewer_mats/run_danpos.sh %s %s %s" % \
(save_file, OUTPUT_DIR, danpos_path)
output, error = run_cmd(bash_command, stdout_file=None)
danpos_calls_path = '%s/result/pooled/mnase_0.smooth.positions.xls' % \
(danpos_output)
danpos_positions = pd.read_csv(danpos_calls_path, sep='\t')
plt.hist(danpos_positions[danpos_positions.smt_value < 10000].smt_value,
bins=100)
plt.savefig("%s/danpos_smt_pos.png" % danpos_output)
danpos_positions = danpos_positions.sort_values('smt_value',
ascending=False)
top_danpos = danpos_positions.head(2500)
top_danpos = top_danpos.rename(columns={
'chr': 'chromosome',
'smt_pos': 'position'
})
from src.chromatin import collect_mnase
from src.kernel_fitter import compute_nuc_kernel
nuc_kernel = compute_nuc_kernel(mnase, top_danpos)
nuc_kernel.save_kernel("%s/danpos_kernel.json" % danpos_output)
from src.kernel_fitter import compute_triple_kernel
nuc_kernel.plot_kernel(kernel_type='nucleosome')
plt.savefig('%s/danpos_nuc_kernel.pdf' % (save_dir), transparent=True)
triple_kernel = compute_triple_kernel(nuc_kernel)
triple_kernel.plot_kernel(kernel_type='triple')
plt.savefig('%s/danpos_triple_kernel.pdf' % (save_dir), transparent=True)
def save_data(self):
save_dir = '%s/tf_analysis' % OUTPUT_DIR
mkdirs_safe([save_dir])
print_fl("Saving %s" % save_dir)
self.all_peaks.to_csv('%s/all_peaks.csv' % save_dir)
self.linked_peaks_normalized.to_csv('%s/linked_peaks_norm.csv' % save_dir)
self.linked_peaks.to_csv('%s/linked_peaks.csv' % save_dir)
self.prom_peaks.to_csv('%s/prom_peaks.csv' % save_dir)
self.all_motifs.to_csv('%s/all_motifs.csv' % save_dir)
def calculate_promoter_regions():
from config import OUTPUT_DIR
all_orfs = all_orfs_TSS_PAS()
gene_boundaries = load_park_orf_transcript_boundaries()
gene_boundaries = gene_boundaries.join(all_orfs[['orf_class', 'name']])
gene_boundaries = gene_boundaries[gene_boundaries.orf_class != 'Dubious']
# find promoter regions using gene boundaries
gene_boundaries = gene_boundaries.sort_values(['chr', 'transcript_start'])
gene_boundaries['promoter_start'] = -1
gene_boundaries['promoter_stop'] = -1
default_prom_size = 1000
for chrom in range(1, 17):
chrom_genes = gene_boundaries[gene_boundaries.chr == chrom]
for idx, gene in chrom_genes.iterrows():
if gene.strand == '+':
prom_start = gene.transcript_start-default_prom_size # default promoter boundary
upstream_genes = chrom_genes[(chrom_genes.transcript_stop > prom_start) &
(chrom_genes.transcript_stop < gene.transcript_start)]
# how close is the closest upstream gene
if len(upstream_genes) > 0:
upstream_genes = upstream_genes.sort_values('transcript_stop', ascending=False)
prom_start = upstream_genes.reset_index().loc[0].transcript_stop
# promoter defined by closest upstream stop and transcript start
gene_boundaries.loc[idx, 'promoter_start'] = prom_start
gene_boundaries.loc[idx, 'promoter_stop'] = gene.transcript_start
elif gene.strand == '-':
prom_start = gene.transcript_stop+default_prom_size # default promoter boundary
upstream_genes = chrom_genes[(chrom_genes.transcript_start < prom_start) &
(chrom_genes.transcript_start > gene.transcript_stop)]
# how close is the closest upstream gene
if len(upstream_genes) > 0:
upstream_genes = upstream_genes.sort_values('transcript_start')
prom_start = upstream_genes.reset_index().loc[0].transcript_start
# promoter defined by closest upstream stop and transcript start
gene_boundaries.loc[idx, 'promoter_stop'] = prom_start
gene_boundaries.loc[idx, 'promoter_start'] = gene.transcript_stop
save_path = '%s/calculated_promoters.csv' % OUTPUT_DIR
gene_boundaries.to_csv(save_path)
print_fl("Saved to %s" % save_path)
return gene_boundaries
def calculate_cross_correlation_all_chromosomes(mnase_seq,
TSSs,
nuc_kernel,
sm_kernel,
triple_kernel,
log=True,
save_chrom_dir=None,
timer=None,
find_antisense=False):
name = task_name(find_antisense)
driver = TaskDriver(name, WATCH_TMP_DIR, 16, timer=timer)
driver.print_driver()
# for all chromosomes calculate occupancies per orf
for chrom in range(1, 17):
if not USE_SLURM:
calculate_cross_correlation_chr(mnase_seq, TSSs, chrom,
find_antisense, nuc_kernel,
sm_kernel, triple_kernel,
save_chrom_dir, log, timer)
child_done(name, WATCH_TMP_DIR, chrom)
else:
exports = (
"CHROM=%d,ANTISENSE=%s,SLURM_WORKING_DIR=%s,CONDA_PATH=%s,CONDA_ENV=%s"
% (chrom, str(find_antisense), SLURM_WORKING_DIR, CONDA_PATH,
CONDA_ENV))
script = 'scripts/2_preprocessing/cross_correlation.sh'
submit_sbatch(exports, script, WATCH_TMP_DIR)
# wait for all chromosomes to finish
# superfluous if not in SLURM mode
driver.wait_for_tasks()
print_fl()
# merge
summary_cross = pd.DataFrame()
cross = pd.DataFrame()
for chrom in range(1, 17):
chrom_cross = pd.read_hdf(cross_filename(save_chrom_dir, chrom),
'cross_correlation')
if len(chrom_cross) == 0: continue
chrom_summary = pd.read_csv(summary_filename(save_chrom_dir, chrom))\
.set_index('orf_name')
cross = cross.append(chrom_cross)
summary_cross = summary_cross.append(chrom_summary)
summary_cross = np.round(summary_cross, 5)
cross = np.round(cross, 5)
return cross, summary_cross
def write_gene_plots(self,
genes,
cc_dir,
lines_dir,
show_plot=True,
custom_lims={},
suffix='',
large_font=False):
for gene_name in genes:
# create heatmaps of the cross correlation for each gene
write_path = "%s/%s%s.pdf" % (cc_dir, gene_name, suffix)
try:
self.set_gene(gene_name)
except KeyError:
print_fl("Could not plot %s" % gene_name)
continue
fig = self.plot_cross_correlation_heatmap(
show_colorbar=True,
title='$\it{' + gene_name + '}$ cross correlation',
large_font=large_font)
plt.savefig(write_path, transparent=False)
# close plots
if not show_plot:
plt.close(fig)
plt.cla()
plt.clf()
if gene_name in custom_lims.keys():
lims = custom_lims[gene_name]
else:
lims = (None, None, None)
# plot lines plots of time course
write_path = "%s/%s%s.pdf" % (lines_dir, gene_name, suffix)
fig = self.plot_lines(self.gene.name,
title=r'$\it{' + gene_name +
'}$ time course',
lims=lims,
large_font=large_font)
plt.savefig(write_path, transparent=False)
# close
if not show_plot:
plt.close(fig)
plt.cla()
plt.clf()
def subset_genes(self, head=None, tail=None):
if head is not None:
subset_func = pd.Series.head
N = head
elif tail is not None:
subset_func = pd.Series.tail
N = tail
self.N = N
data = self.agg_fun(self.gene_body_disorganization_delta, axis=1)
self.cur_disorg_orfs = subset_func(data.sort_values(ascending=False), N).index.values
data = self.agg_fun(self.promoter_sm_occupancy_delta, axis=1)
self.cur_promoter_orfs = subset_func(data.sort_values(ascending=False), N).index.values
data = self.agg_fun(self.chromatin_data, axis=1)
self.cur_chromatin_orfs = subset_func(data.sort_values(ascending=False), N).index.values
data = self.agg_fun(self.xrate_data, axis=1)
self.cur_xrate_orfs = subset_func(data.sort_values(ascending=False), N).index.values
# TODO:
# Report the quantile of the selection
print_fl("Disorganization ORFs: %d" % len(self.cur_disorg_orfs))
print_fl("Promoter ORFs: %d" % len(self.cur_promoter_orfs))
print_fl("Chromatin ORFs: %d" % len(self.cur_chromatin_orfs))
print_fl("Transcription ORFs: %d" % len(self.cur_xrate_orfs))
def sra_download_convert_bam(write_dir, sra_id, filename):
prefetch = "%s/prefetch" % SRA_BIN_DIR
sam_dump = "%s/sam-dump" % SRA_BIN_DIR
# prefetch SRA ID
print_fl("Prefetching %s" % sra_id)
sra_write_path = "%s/%s.sra" % (write_dir, filename)
run_cmd("%s %s --output-file %s" % (prefetch, sra_id, sra_write_path))
# dump to sam
sam_write_path = "%s/%s.sam" % (write_dir, filename)
print_fl("Dumping SAM %s" % sam_write_path)
run_cmd("%s %s" % (sam_dump, sra_write_path), stdout_file=sam_write_path)
# convert to bam
bam_write_path = "%s/%s.bam" % (write_dir, filename)
print_fl("Converting to BAM %s" % bam_write_path)
run_cmd("samtools view -b -S %s" % (sam_write_path),
stdout_file=bam_write_path)
# index
bam_write_path = "%s/%s.bam" % (write_dir, filename)
bam_index_path = "%s/%s.bam.bai" % (write_dir, filename)
print_fl("Indexing BAM %s" % bam_index_path)
run_cmd("samtools index %s %s" % (bam_write_path, bam_index_path))
# remove large SAM file
os.remove(sam_write_path)
def call_nucleosomes_p123_chrom(orfs, chrom, antisense,
cross_correlation_dir, save_chrom_dir, timer):
chrom_orfs = orfs[orfs.chr == chrom]
if len(chrom_orfs) == 0: return None
print_fl("Chromosome %d. %d genes" % (chrom, len(chrom_orfs)))
timer.print_time()
p123_orfs = chrom_orfs[[]].copy()
p123_orfs['+1'] = np.nan
p123_orfs['+2'] = np.nan
p123_orfs['+3'] = np.nan
linkages = pd.DataFrame()
# load relevant cross correlations
chrom_cross_correlation = pd.read_hdf(
'%s/cross_correlation_chr%d.h5.z' %
(cross_correlation_dir, chrom))
i = 0
for idx, orf in chrom_orfs.iterrows():
if i % 200 == 0:
print_fl(" %d/%d - %s" % (i, len(chrom_orfs), timer.get_time()))
i += 1
# call nucleosomes and link them in ORF window
try:
nucs = call_orf_nucleosomes(chrom_cross_correlation.loc['diff'],
orf)
cur_linkages = find_linkages(nucs)
linkages = linkages.append(cur_linkages)
except KeyError:
continue
# +1, +2, +3
p1, p2, p3 = find_p123_gene(orf, linkages)
p123_orfs.loc[idx] = [p1, p2, p3]
save_path = nucleosomes_filename(save_chrom_dir, chrom)
linkages.to_csv(save_path)
p123_orfs = p123_orfs.dropna()
save_path = p123_filename(save_chrom_dir, chrom)
p123_orfs.to_csv(save_path)
def fit_position(self):
"""Fit positional distribution to a normal with uniform background"""
Y = self.pivoted_data.sum(axis=0)
# fixed parameters, mean of normal and uniform range
mean_position = 0
if self.kernel_type == 'small':
self.pos_std = np.std(Y)
else:
self.pos_std = np.std(Y)
self.pos_mean = mean_position
print_fl("Kernel positional mean %.2f and std %.2f" %
(self.pos_mean, self.pos_std))
def compute_nuc_kernel(all_mnase_data, brogaard, top=2500):
# Get MNase-seq data @ 0 min for top 2500 nucleosomes
mnase_seq_0 = filter_mnase(all_mnase_data, time=0.0)
top_brogaard = brogaard.head(top)
brogaard_mnase = collect_mnase(mnase_seq_0, window=200, pos_chr_df=top_brogaard)
nuc_length_mode = int(brogaard_mnase.length.mode())
print_fl("Nucleosome length mode: %d" % nuc_length_mode)
nuc_fitter = KernelFitter(brogaard_mnase, len_mean=nuc_length_mode, window=200, kernel_type='nucleosome')
nuc_fitter.fit()
nuc_fitter.generate_kernel()
return nuc_fitter.kernel
def compute_sm_kernel(all_mnase_data, abf1_sites):
mnase_seq_0 = filter_mnase(all_mnase_data, time=0.0)
abf1_mnase = collect_mnase(mnase_seq_0, window=150,
pos_chr_df=abf1_sites, chrom_key='chr',
pos_key='mid',
strand='strand')
abf1_length_mode = int(abf1_mnase.length.mode())
print_fl("Abf1 fragment length mode: %d" % abf1_length_mode)
sm_fitter = KernelFitter(abf1_mnase, len_mean=abf1_length_mode, window=150,
kernel_type='small')
sm_fitter.fit()
sm_fitter.generate_kernel()
return sm_fitter.kernel
def fit_length(self):
Y = self.pivoted_data.sum(axis=1)
if self.kernel_type == 'small':
# mirror small fragments distribution from 0-mode to compute standard deviation in length
data = self.pivoted_data.sum(axis=1).loc[0:self.len_mean]
mirrored_data = data.sort_index(ascending=False)
data = np.concatenate([data, mirrored_data])
self.len_std = np.std(data)#/50.
elif self.kernel_type == 'nucleosome':
data = self.pivoted_data.sum(axis=1).loc[self.len_mean:]
mirrored_data = data.sort_index(ascending=False)
data = np.concatenate([mirrored_data, data])
self.len_std = np.std(data)#/500.
print_fl("Kernel %s length mean %.2f and std %.2f" %
(self.kernel_type, self.len_mean, self.len_std), log=True)
def fit(self, k=10):
with TimingContext() as timing:
for model in self.models:
print_fl("Fitting %s" % model.name)
model.fit_cv(log=False, k=k)
print_fl(" " + timing.get_time())
times = model.times
df = pd.DataFrame(index=times)
for model in self.models:
df[model.name] = model.mse
self.mse = df
df = pd.DataFrame(index=times)
for model in self.models:
df[model.name] = model.r2
self.r2 = df
def call_all_nucleosome_p123(orfs, antisense,
cross_correlation_dir, chrom_save_dir, timer):
linkages = pd.DataFrame()
name = task_name(antisense)
driver = TaskDriver(name, WATCH_TMP_DIR, 16, timer=timer)
driver.print_driver()
for chrom in range(1, 17):
if not USE_SLURM:
call_nucleosomes_p123_chrom(orfs, chrom, antisense,
cross_correlation_dir, chrom_save_dir, timer)
child_done(name, WATCH_TMP_DIR, chrom)
else:
exports = ("CHROM=%d,ANTISENSE=%s,SLURM_WORKING_DIR=%s,CONDA_PATH=%s,CONDA_ENV=%s"
% (chrom, str(antisense), SLURM_WORKING_DIR, CONDA_PATH, CONDA_ENV))
script = 'scripts/2_preprocessing/call_nucleosomes.sh'
submit_sbatch(exports, script, WATCH_TMP_DIR)
# wait for all chromosomes to finish
# superfluous if not in SLURM mode
driver.wait_for_tasks()
print_fl()
# merge
nucleosomes = pd.DataFrame()
p123 = pd.DataFrame()
for chrom in range(1, 17):
if not os.path.exists(nucleosomes_filename(chrom_save_dir, chrom)):
continue
nuc_chr = pd.read_csv(nucleosomes_filename(chrom_save_dir, chrom))\
.set_index('orf')
p123_chr = pd.read_csv(p123_filename(chrom_save_dir, chrom))\
.set_index('orf_name')
nucleosomes = nucleosomes.append(nuc_chr)
p123 = p123.append(p123_chr)
return nucleosomes, p123
def __init__(self,
name,
times=[0, 7.5, 15, 30, 60, 120],
sample_N=None,
results_path=None):
print_fl("Loading %s" % name)
self.name = name
self.sample_N = sample_N
self.times = times
if results_path is None:
self.design_matrix()
else:
self.load_results(results_path)
self.l_scale = 1
self.l_bounds = 1, 10
def main():
print_fl("Loading models")
gp_compare = RegressionCompare(reg_model=GP)
gp_compare.fit(k=10)
gp_compare.plot_compare(metric='r2')
plt.savefig('output/gp/r2.pdf', transparent=True)
gp_compare.plot_compare(metric='mse')
plt.savefig('output/gp/mse.pdf', transparent=True)
gp_compare.full_model.plot_fit()
plt.savefig('output/gp/full.pdf', transparent=True)
gp_compare.full_model.plot_fit(120)
plt.savefig('output/gp/full_120.pdf', transparent=True)
gp_compare.mse.T.to_csv('output/gp/model_mse.csv', float_format='%.4f')
gp_compare.r2.T.to_csv('output/gp/model_r2.csv', float_format='%.4f')
def main():
print_fl("*******************************")
print_fl("* 6 Reviewer Materials *")
print_fl("*******************************")
print_preamble()
mkdirs_safe([save_dir])
plot_utils.apply_global_settings()
# plots for shift edge analysis
shift_edge_analysis.main()
# additional scatter plots
scatters()
xrate_vs_TPM()
# danpos
danpos()
# OD curve
plot_OD_curve()
def run_models(save_dir, timer):
task_name = 'gp'
# launch gp models
print_fl("Loading models...", end='')
models = get_model_funs()
print_fl("Running %d models..." % len(models), end='')
driver = TaskDriver(task_name,
WATCH_TMP_DIR,
len(models.keys()),
timer=timer)
driver.print_driver()
for name, model in models.items():
if not USE_SLURM:
run_model(name, save_dir)
child_done(task_name, WATCH_TMP_DIR, name)
pass
else:
exports = ("MODEL=%s,SLURM_WORKING_DIR=%s,CONDA_PATH=%s,CONDA_ENV=%s" % \
(name.replace(' ', '_'),
SLURM_WORKING_DIR, CONDA_PATH, CONDA_ENV))
script = 'scripts/4_analysis/gp.sh'
submit_sbatch(exports, script, WATCH_TMP_DIR)
driver.wait_for_tasks()
print_fl()
def compute_cross_correlations(strand='sense'):
from src.cross_correlation_kernel import MNaseSeqDensityKernel
from src.cross_correlation import calculate_cross_correlation_all_chromosomes
cc_orfs = paper_orfs
cc_dir = cc_sense_chrom_dir
cross_corr_path = cross_corr_sense_path
if strand == 'antisense':
cc_orfs = antisense_orfs
cc_dir = cc_antisense_chrom_dir
cross_corr_path = cross_corr_antisense_path
mkdirs_safe([cc_dir])
nuc_kernel = MNaseSeqDensityKernel(filepath=nuc_kernel_path)
sm_kernel = MNaseSeqDensityKernel(filepath=sm_kernel_path)
triple_kernel = compute_triple_kernel(nuc_kernel)
print_fl("Cross correlating %d ORFs..." % len(cc_orfs))
cross, summary_cross = calculate_cross_correlation_all_chromosomes(
all_mnase_data, cc_orfs, nuc_kernel, sm_kernel, triple_kernel,
save_chrom_dir=cc_dir, timer=timer, log=True,
find_antisense=(strand == 'antisense'))
cross.to_hdf(cross_corr_path,
'cross_correlation', mode='w', complevel=9, complib='zlib')
summary_cross.to_csv('%s/cross_correlation_summary_%s.csv' %
(mnase_dir, strand))
print_fl("Done.")
timer.print_time()
print_fl()
def create_merged_metrics_df(write_dir, occupancy, cross):
if cross:
# merge cross correlation dataframes
print_fl("Merging cross correlation...")
cross_correlation = pd.DataFrame()
for chrom in range(1, 17):
path = "%s/orf_cross_correlation_chr%d.h5.z" % (write_dir, chrom)
cur_df = pd.read_hdf(path, 'cross_correlation')
cross_correlation = cross_correlation.append(cur_df)
# write to disk
cross_correlation.to_hdf('%s/orf_cross_correlation.h5.z' % write_dir,
'cross_correlation',
mode='w',
complevel=9,
complib='zlib')
# merge cross correlation summary
print_fl("Merging cross correlation summary...")
orf_cross_correlation_summary = pd.DataFrame()
for chrom in range(1, 17):
path = "%s/orf_cross_correlation_summary_chr%d.csv" % (write_dir,
chrom)
cur_df = pd.read_csv(path)
orf_cross_correlation_summary = orf_cross_correlation_summary.append(
cur_df)
# write to disk
orf_cross_correlation_summary.to_csv(
'%s/orf_cross_correlation_summary.csv' % write_dir, index=False)
# merge coverage
if occupancy:
print_fl("Merging coverage...")
coverage = pd.DataFrame()
for chrom in range(1, 17):
path = "%s/coverage_chr%d.csv" % (write_dir, chrom)
cur_df = pd.read_csv(path)
coverage = coverage.append(cur_df)
# write to disk
coverage.to_csv('%s/coverage.csv' % write_dir, index=False)
# merge coverage
print_fl("Merging occupancy...")
occupancy = pd.DataFrame()
for chrom in range(1, 17):
path = "%s/occupancy_chr%d.csv" % (write_dir, chrom)
cur_df = pd.read_csv(path)
occupancy = occupancy.append(cur_df)
# write to disk
occupancy.to_csv('%s/occupancy.csv' % write_dir, index=False)
def get_cross_correlation(wide_counts_df,
kernel,
times=[0.0, 7.5, 15, 30, 60, 120]):
"""
Assumes ndarray of (orf, time, length, position)
"""
# calculate indices to be create the resulting dataframe
kernel_span = kernel.extent[0], kernel.extent[1]
positions = wide_counts_df.columns.values
pos_span = positions.min(), positions.max()
kernel_width_2 = (kernel_span[1] - kernel_span[0]) / 2
result_span = pos_span[0] + kernel_width_2, pos_span[1] - kernel_width_2
result_len = result_span[1] - result_span[0] + 1
orf_idxs = wide_counts_df.index.levels[0]
n = 1
num_times = len(times)
kern_mat = kernel.kernel_mat
conv_df = create_orfs_time_df(orf_idxs,
columns=np.arange(result_span[0],
result_span[1] + 1))
for orf_name in orf_idxs:
for time in times:
try:
cur_arr = wide_counts_df.loc[orf_name].loc[time].values
except Exception as e:
print_fl("Exception thrown for ORF %s.\n%s" %
(orf_name, str(e)))
continue
cur_conv_score = correlate2d(cur_arr, kern_mat, mode='valid')
conv_df.loc[orf_name].loc[time] = cur_conv_score
return conv_df.astype(float)
def print_go(self):
go_terms = {}
for cluster in self.clusters:
terms = list(self.clustered_go_sig[\
self.clustered_go_sig['cluster'] == cluster]['name'].values)
# remove high-level terms
drop_items = {
'molecular_function', 'cytoplasm', 'cellular_component',
'biological_process', 'nucleolus', 'cytoplasmic vesicle'
}
terms = set(terms) - drop_items
terms = [t[0].upper() + t[1:] for t in terms]
go_value = ('Cluster %d\n' % (cluster)) + '\n'.join(list(terms))
go_terms[str(cluster)] = go_value
if len(terms) > 0:
print_fl(go_value)
print_fl("-------------------")
self.go_terms = go_terms
def __init__(self,
mnase_path=None,
rna_seq_pileup_path=None,
orfs=None,
times=[0.0, 7.5, 15, 30, 60, 120]):
self.orfs = orfs
print_fl("Loading MNase-seq...")
self.all_mnase_data = pd.read_hdf(mnase_path, 'mnase_data')
self.CDS_introns = read_sgd_orf_introns()
print_fl("Loading RNA-seq pileup...")
pileup = pd.read_hdf(rna_seq_pileup_path, 'pileup')
self.rna_seq_plotter = RNASeqPlotter(pileup)
self.orfs_plotter = ORFAnnotationPlotter(orfs, self.CDS_introns)
self.times = times
self.span = None
self.chrom = None
self.set_config()