def filter(p_bed, region_bed, max_p=None, p_col_name="P.Value"):
ph = ['p' + h for h in get_header(p_bed)]
rh = get_header(region_bed)
if isinstance(p_col_name, (int, long)):
p_col_name = ph[p_col_name][1:]
a = dict(p_bed=p_bed, region_bed=region_bed)
a['p_bed'] = fix_header(a['p_bed'])
yield rh + ["t-pos", "t-neg", "t-sum", "n_gt_p05", "n_gt_p1"]
for group, plist in groupby(reader('|bedtools intersect -b %(p_bed)s -a %(region_bed)s -wo' % a,
header=rh + ph), itemgetter('chrom','start','end')):
plist = list(plist)
plist = [x for x in plist if (int(x['start']) <= int(x['pstart']) <= int(x['pend'])) and ((int(x['start']) <= int(x['pend']) <= int(x['end'])))]
tscores = [float(row['pt']) for row in plist if 'pt' in row]
if max_p:
if any(float(row['p' + p_col_name]) > max_p for row in plist):
continue
ngt05 = sum(1 for row in plist if float(row['p' + p_col_name]) > 0.05)
ngt1 = sum(1 for row in plist if float(row['p' + p_col_name]) > 0.1)
tpos = sum(1 for ts in tscores if ts > 0)
tneg = sum(1 for ts in tscores if ts < 0)
tsum = sum(ts for ts in tscores)
frow = [plist[0][h] for h in rh] + [str(tpos), str(tneg), str(tsum), str(ngt05), str(ngt1)]
yield frow
def partsort(afile, group_cols, sort_cols, sort_convertors, header=False):
"""
the converted columns are appended to the end of the row.
then after the sort, these are removed.
this removes problems with floating point reprs.
"""
the_first_line = get_header(afile)
row_len = len(the_first_line)
n_extra = len(sort_convertors)
# maintain order of the sort cols, but use the appended columns for the
# numeric ones.
actual_sort_cols = []
n_extra = 0
# since we append floats to the end *and* want to maintain the
# requested sort order, we create the `actual_sort_cols`
for c in sort_cols:
if not c in sort_convertors:
actual_sort_cols.append(c)
else:
idx = row_len + n_extra
actual_sort_cols.append(idx)
n_extra += 1
# if it was stdin, then we read one line to get the header length.
lines = reader(afile, header=header) if afile != "-" \
else chain([the_first_line], reader(afile, header))
# groupby the correct columns
for keyed, group in groupby(lines,
lambda toks: [toks[i] for i in group_cols]):
# then generate the rows with the converted columns appended.
def gen_converted_group():
for toks in group:
# add the converted columns onto the end.
yield toks + [
fn(toks[col_idx])
for col_idx, fn in sort_convertors.items()
]
# then iterator over the sorted cols.
for toks in sorted(gen_converted_group(),
key=itemgetter(*actual_sort_cols)):
# strip the extra columns.
yield toks[:row_len]
def partsort(afile, group_cols, sort_cols, sort_convertors, header=False):
"""
the converted columns are appended to the end of the row.
then after the sort, these are removed.
this removes problems with floating point reprs.
"""
the_first_line = get_header(afile)
row_len = len(the_first_line)
n_extra = len(sort_convertors)
# maintain order of the sort cols, but use the appended columns for the
# numeric ones.
actual_sort_cols = []
n_extra = 0
# since we append floats to the end *and* want to maintain the
# requested sort order, we create the `actual_sort_cols`
for c in sort_cols:
if not c in sort_convertors:
actual_sort_cols.append(c)
else:
idx = row_len + n_extra
actual_sort_cols.append(idx)
n_extra += 1
# if it was stdin, then we read one line to get the header length.
lines = reader(afile, header=header) if afile != "-" \
else chain([the_first_line], reader(afile, header))
# groupby the correct columns
for keyed, group in groupby(lines, lambda toks:
[toks[i] for i in group_cols]):
# then generate the rows with the converted columns appended.
def gen_converted_group():
for toks in group:
# add the converted columns onto the end.
yield toks + [fn(toks[col_idx]) for col_idx, fn in sort_convertors.items()]
# then iterator over the sorted cols.
for toks in sorted(gen_converted_group(), key=itemgetter(*actual_sort_cols)):
# strip the extra columns.
yield toks[:row_len]
def filter(p_bed, region_bed, max_p=None, region_p=None, p_col_name="P.Value",
coef_col_name="logFC"):
ph = ['p' + h for h in get_header(p_bed)]
rh = get_header(region_bed)
if isinstance(p_col_name, (int, long)):
p_col_name = ph[p_col_name][1:]
a = dict(p_bed=p_bed, region_bed=region_bed)
a['p_bed'] = fix_header(a['p_bed'])
j = 0
for group, plist in groupby(reader('|bedtools intersect -b %(p_bed)s -a %(region_bed)s -wo' % a,
header=rh + ph), itemgetter('chrom','start','end')):
plist = list(plist)
if region_p:
r = plist[0] # first cols are all the same
region_p_key = 'slk_sidak_p' if 'slk_sidak_p' in r \
else 'z_sidak_p' if 'z_sidak_p' in r \
else None
if region_p_key is None: raise Exception
if float(r[region_p_key]) > region_p:
continue
plist = [x for x in plist if (int(x['start']) <= int(x['pstart']) <= int(x['pend'])) and ((int(x['start']) <= int(x['pend']) <= int(x['end'])))]
tscores = [float(row['pt']) for row in plist if 'pt' in row]
if max_p:
if any(float(row['p' + p_col_name]) > max_p for row in plist):
continue
ngt05 = sum(1 for row in plist if float(row['p' + p_col_name]) > 0.05)
# logic to try to find t and coef headers and skip if not found
extra_header = []
extra = []
if tscores:
tpos = sum(1 for ts in tscores if ts > 0)
tneg = sum(1 for ts in tscores if ts < 0)
tpn = "%i/%i" % (tpos, tneg)
tsum = str(sum(ts for ts in tscores))
extra_header += ["t.pos/t.neg", "t.sum"]
extra += [tpn, tsum]
else:
tsum = tpn = "NA"
if 'p' + coef_col_name not in plist[0] and 'pcoefficient' in plist[0]:
coef_col_name = 'coefficient'
if 'p' + coef_col_name in plist[0]:
coef = (sum(float(row['p' + coef_col_name]) for row in plist) /
len(plist))
# since we probably had the data logit transformed, here we
# do the inverse and subtract 0.5 since ilogit(0) == 0.5
icoef = (sum(ilogit(float(row['p' + coef_col_name])) for row in plist) /
len(plist)) - 0.5
extra_header += ["avg.diff", "ilogit.diff"]
extra += ["%.3f" % coef, "%.3f" % icoef]
else:
coef = icoef = float('nan')
frow = [plist[0][h] for h in rh] + extra
if j == 0:
yield rh + extra_header
j = 1
yield frow