def estimate_window_size(readData, parameter):
if (parameter.window_size != -1):
# window size should be even number
if parameter.window_size % 2 ==1:
info("warning: window size should be even number; adding 1 to it.")
parameter.window_size = parameter.window_size + 1
return parameter.window_size
info( "begin window_size calculation...")
window_size_list = []
for chr in readData.chr_list:
reads = []
for chip_filename in readData.chip_filename_list:
reads.extend(readData.data_dict[chr][chip_filename])
coord_array = numpy.zeros(readData.chr_length_dict[chr],
dtype=numpy.int64)
for x in reads:
try: coord_array[x] += 1
except IndexError: pass #debug("coordinates out of range")
window = get_window_size2(coord_array)
# if window >1000: # limit the window size to be less than 1000 bases.
# window = 1000
window_size_list.append(window)
window_logger.debug("%-10s %s", chr, window)
debug("length of windowsize list is "+str(len(window_size_list)))
window_median = misc.median(window_size_list)
# Window size cannot be odd number since we need
# to use overlapping half windows.
if window_median % 2 == 1:
window_median = window_median+1
window_logger.debug("window_size: %+10s", window_median)
info ( "finishing window size calculation...")
parameter.window_size = window_median
return
def estimate_window_size(readData, parameter):
if (parameter.window_size != -1):
# window size should be even number
if parameter.window_size % 2 == 1:
info("warning: window size should be even number; adding 1 to it.")
parameter.window_size = parameter.window_size + 1
return parameter.window_size
info("begin window_size calculation...")
window_size_list = []
for chr in readData.chr_list:
reads = []
for chip_filename in readData.chip_filename_list:
reads.extend(readData.data_dict[chr][chip_filename])
coord_array = numpy.zeros(readData.chr_length_dict[chr],
dtype=numpy.int64)
for x in reads:
try:
coord_array[x] += 1
except IndexError:
pass #debug("coordinates out of range")
window = get_window_size2(coord_array)
# if window >1000: # limit the window size to be less than 1000 bases.
# window = 1000
window_size_list.append(window)
window_logger.debug("%-10s %s", chr, window)
debug("length of windowsize list is " + str(len(window_size_list)))
window_median = misc.median(window_size_list)
# Window size cannot be odd number since we need
# to use overlapping half windows.
if window_median % 2 == 1:
window_median = window_median + 1
window_logger.debug("window_size: %+10s", window_median)
info("finishing window size calculation...")
parameter.window_size = window_median
return
def post_processing_per_peak(strands_dict, chip_list, input_list, chr,
start, end, shiftSize, readLength, narrow_peak,
remove_artefacts):
''' Remove artefacts and refine peak width.'''
chip_forward = numpy.zeros(end-start)
chip_reverse = numpy.zeros(end-start)
input_forward = numpy.zeros(end-start)
input_reverse = numpy.zeros(end-start)
for chip in chip_list:
forward = strands_dict[chr][chip]['f']
reverse = strands_dict[chr][chip]['r']
forward_read = forward[numpy.where( (forward >= start) &
(forward < end) )]
reverse_read = reverse[numpy.where( (reverse >= start+readLength) &
(reverse < end+readLength) )]
for read in forward_read:
try: chip_forward[(read-start)] +=1
except IndexError:
debug("index error ignored. end-start: %d. read-start: %d.", end-start, read-start)
for read in reverse_read:
try: chip_reverse[(read-start-readLength)] +=1
except IndexError:
debug("index error ignored. end-start: %d. read-start-readLength: %d.", end-start, read-start-readLength)
# using input reads to remove artefacts
if remove_artefacts is True:
for input in input_list:
forward = strands_dict[chr][input]['f']
reverse = strands_dict[chr][input]['r']
forward_read = forward[numpy.where( (forward >=
start) & (forward < end) )]
reverse_read = reverse[numpy.where( (reverse >=
start+readLength) & (reverse < end+readLength) )]
for read in forward_read:
try: input_forward[(read-start)] +=1
except IndexError:
debug("index error ignored. end-start: %d. read-start: %d.", end-start, read-start)
for read in reverse_read:
try: input_reverse[(read-start-readLength)] +=1
except IndexError:
debug("index error ignored. end-start: %d. read-start-readLength: %d.", end-start, read-start-readLength)
chip_both = chip_forward + chip_reverse
input_both = input_forward + input_reverse
if sum(chip_both) == 0:
pass
else:
chip_both = chip_both/sum(chip_both)
if sum(input_both) ==0:
pass
else:
input_both = input_both/sum(input_both)
overlap_chip_input = numpy.sum(numpy.minimum(chip_both, input_both))
chip_both.sort()
chip_3_maximum = numpy.sum(chip_both[-3:])
if sum(chip_reverse) != 0:
chip_reverse = chip_reverse/sum(chip_reverse)
chip_forward_roll = numpy.roll(chip_forward,readLength)
if sum(chip_forward) != 0:
chip_forward = chip_forward/sum(chip_forward)
chip_forward_roll = chip_forward_roll/sum(chip_forward)
overlap_orig = numpy.sum(numpy.minimum(chip_forward, chip_reverse))
overlap_orig = numpy.max([overlap_orig, 1e-5])
overlap_roll = numpy.sum(numpy.minimum(chip_forward_roll, chip_reverse))
else:
overlap_chip_input = 0
chip_3_maximum = 0
overlap_orig = 1e-5
overlap_roll = 0
if narrow_peak is True:
sum_forward = 0
sum_reverse = 0
if sum(chip_forward) > 0:
chip_forward = chip_forward/sum(chip_forward)
for i in range(end-start):
sum_forward += chip_forward[i]
if sum_forward > 0.2:
new_start = start + i
break
if sum(chip_reverse) > 0:
chip_reverse = chip_reverse/sum(chip_reverse)
for i in range(end-start-1, -1, -1):
sum_reverse += chip_reverse[i]
if sum_reverse > 0.2:
new_end = start + i
break
# there is no reads in the peak.
if sum(chip_forward)+sum(chip_reverse) == 0:
new_start = new_end = (start+end)/2
if sum(chip_forward) == 0:
new_start = new_end - misc.median(2*shiftSize.values())
if sum(chip_reverse) == 0:
new_end = new_start + misc.median(2*shiftSize.values())
start = new_start
end = new_end
return (start, end, overlap_chip_input, chip_3_maximum,
overlap_orig, overlap_roll)
def get_window_size2(array, bin=20, iter=100):
# this is the function that estimate the window size.
chr_len = array.size
rowNum = int(chr_len/bin)
array_window = std.as_strided(array, (rowNum, bin),
(bin*array.itemsize, 1*array.itemsize))
array_window = numpy.sum(array_window, 1)
peak_len_list = []
for x in range(iter):
peak = numpy.max(array_window)
peak_idx = numpy.where(array_window == peak)[0][0]
# set the peak to -1 so that it won't confuse the next iteration.
array_window[peak_idx]= -1
if peak_idx == 0: # Check if the window reaches the left boundary.
left_boundary_not_reached = False
else:
left_boundary_not_reached = True
# Check if the window reaches of the right boundary.
if peak_idx == rowNum-1:
right_boundary_not_reached = False
else:
right_boundary_not_reached = True
i_l = 1
i_r = 1
# Be careful if anything equals to -1.
# Though it's unlikely that it will happen.
while left_boundary_not_reached:
while (array_window[peak_idx-i_l] >= 0.1*peak):
array_window[peak_idx-i_l] = -1
if peak_idx-i_l==0:
left_boundary_not_reached = False
break
i_l += 1
if left_boundary_not_reached:
if peak_idx-i_l==0:
array_window[peak_idx-i_l] = -1
left_boundary_not_reached = False
if left_boundary_not_reached:
# Will continue counting if the next window(one window gap)
# has above the 10% mode reads.
if (array_window[peak_idx-i_l-1] >= 0.1*peak):
array_window[peak_idx-i_l] = -1
i_l += 1
else:
break
while right_boundary_not_reached:
while (array_window[peak_idx+i_r] >= 0.1*peak):
array_window[peak_idx+i_r] = -1
if peak_idx+i_r ==rowNum-1:
right_boundary_not_reached = False
break
i_r += 1
if right_boundary_not_reached:
if peak_idx+i_r==rowNum-1:
array_window[peak_idx+i_r] = -1
right_boundary_not_reached = False
if right_boundary_not_reached:
if (array_window[peak_idx+i_r+1] >=0.1*peak):
array_window[peak_idx+i_r] = -1
i_r += 1
else:
break
peak_len_list.append((1+i_l+i_r)*bin)
return misc.median(peak_len_list)
def estimate_shift_size(readData, parameter):
# If shift size provided by the user, then skip estimating it.
if parameter.shift_size != "-1":
shift_list = parameter.shift_size.split(',')
if len(shift_list)== 1:
for name in readData.filename_list:
readData.shift_size[name] = int(shift_list[0])
info("%-10s %s", name, shift_list[0])
else:
if parameter.difftest is True:
for idx, name in enumerate(readData.chip_filename_list):
readData.shift_size[name] = int(shift_list[idx])
info("%-10s %s", name, shift_list[idx])
chip1_shift_list = [readData.shift_size[chip]
for chip in readData.chip1_filename_list]
input1_shift_size = sum(chip1_shift_list)/len(chip1_shift_list)
for name in readData.input1_filename_list:
readData.shift_size[name] = input1_shift_size
chip2_shift_list = [readData.shift_size[chip]
for chip in readData.chip2_filename_list]
input2_shift_size = sum(chip2_shift_list)/len(chip2_shift_list)
for name in readData.input2_filename_list:
readData.shift_size[name] = input2_shift_size
else:
for idx, chip in enumerate(readData.chip1_filename_list):
readData.shift_size[chip] = int(shift_list[idx])
info("%-10s %s", chip, shift_list[idx])
for input in readData.input1_filename_list:
readData.shift_size[input] = \
sum([int(x) for x in shift_list])/len(shift_list)
info("%-10s %s", input,
sum([int(x) for x in shift_list])/len(shift_list))
return
info("begin estimating the shift size...")
for chip_filename in readData.chip1_filename_list:
info("estimating for %-10s", chip_filename)
shift_list = []
for count, chr in enumerate(readData.chr_list):
# estimate the shift size for five chromosomes and take
# the median of these as the estimator.
if count ==5:
break
forward = readData.data_dict_by_strands[chr][chip_filename]['f']
reverse = readData.data_dict_by_strands[chr][chip_filename]['r']
shift = shift_size_per_chrom(forward, reverse)
info("%-10s %d", chr, shift)
shift_list.append(shift)
shift_median = misc.median(shift_list[:])
info("%-10s %d", chip_filename, shift_median)
readData.shift_size[chip_filename] = shift_median
chip1_shift_list = [readData.shift_size[chip]
for chip in readData.chip1_filename_list]
input1_shift_size = sum(chip1_shift_list)/len(chip1_shift_list)
for input_filename in readData.input1_filename_list:
readData.shift_size[input_filename] = input1_shift_size
if parameter.difftest is True: # If we're calling differential binding.
for filename in readData.chip2_filename_list:
info("estimating for %-10s", filename)
shift_list = []
for count, chr in enumerate(readData.chr_list):
if count==5:
break
forward = readData.data_dict_by_strands[chr][filename]['f']
reverse = readData.data_dict_by_strands[chr][filename]['r']
shift = shift_size_per_chrom(forward, reverse)
info("%-10s %d", chr, shift)
shift_list.append(shift)
shift_median = misc.median(shift_list[:])
info("%-10s %d", filename, shift_median)
readData.shift_size[filename] = shift_median
chip2_shift_list = [readData.shift_size[chip]
for chip in readData.chip2_filename_list]
input2_shift_size = sum(chip2_shift_list)/len(chip2_shift_list)
for input_filename in readData.input2_filename_list:
readData.shift_size[input_filename] = input2_shift_size
return
def get_window_size2(array, bin=20, iter=100):
# this is the function that estimate the window size.
chr_len = array.size
rowNum = chr_len / bin
array_window = std.as_strided(array, (rowNum, bin),
(bin * array.itemsize, 1 * array.itemsize))
array_window = numpy.sum(array_window, 1)
peak_len_list = []
for x in range(iter):
peak = numpy.max(array_window)
peak_idx = numpy.where(array_window == peak)[0][0]
# set the peak to -1 so that it won't confuse the next iteration.
array_window[peak_idx] = -1
if peak_idx == 0: # Check if the window reaches the left boundary.
left_boundary_not_reached = False
else:
left_boundary_not_reached = True
# Check if the window reaches of the right boundary.
if peak_idx == rowNum - 1:
right_boundary_not_reached = False
else:
right_boundary_not_reached = True
i_l = 1
i_r = 1
# Be careful if anything equals to -1.
# Though it's unlikely that it will happen.
while left_boundary_not_reached:
while (array_window[peak_idx - i_l] >= 0.1 * peak):
array_window[peak_idx - i_l] = -1
if peak_idx - i_l == 0:
left_boundary_not_reached = False
break
i_l += 1
if left_boundary_not_reached:
if peak_idx - i_l == 0:
array_window[peak_idx - i_l] = -1
left_boundary_not_reached = False
if left_boundary_not_reached:
# Will continue counting if the next window(one window gap)
# has above the 10% mode reads.
if (array_window[peak_idx - i_l - 1] >= 0.1 * peak):
array_window[peak_idx - i_l] = -1
i_l += 1
else:
break
while right_boundary_not_reached:
while (array_window[peak_idx + i_r] >= 0.1 * peak):
array_window[peak_idx + i_r] = -1
if peak_idx + i_r == rowNum - 1:
right_boundary_not_reached = False
break
i_r += 1
if right_boundary_not_reached:
if peak_idx + i_r == rowNum - 1:
array_window[peak_idx + i_r] = -1
right_boundary_not_reached = False
if right_boundary_not_reached:
if (array_window[peak_idx + i_r + 1] >= 0.1 * peak):
array_window[peak_idx + i_r] = -1
i_r += 1
else:
break
peak_len_list.append((1 + i_l + i_r) * bin)
return misc.median(peak_len_list)
def check_sadi_consistence(self, fragment):
"""
check if same distance restraints make sense. Each length of an atom
pair is tested agains the standard deviation of all distances.
For a large standard deviation, the list is tested for outliers.
:param fragment: frag name
"""
atoms = self.get_atoms(fragment)
restr = self.get_restraints(fragment)
restraints = deepcopy(restr)
atnames = self.get_atomnames(fragment, uppercase=True)
good = True
for num, line in enumerate(restraints):
prefixes = []
dev = 0.02
line = line.split()
if not line:
continue
if line[0].upper() == 'SADI':
prefixes.append(line[0])
del line[0]
try:
if not str(line[0][0]).isalpha():
prefixes.append(line[0])
dev = line[0]
del line[0] # delete standard deviation
except IndexError:
return False
if len(line) % 2 == 1: # test for uneven atoms count
print('*** Inconsistent SADI restraint line {} of "{}". '
'Not all atoms form a pair ***'.format(num, fragment))
pairs = pairwise(line)
distances = []
pairlist = []
if len(pairs) <= 2:
return True
for i in pairs:
if i in pairlist or tuple(reversed(i)) in pairlist:
print('*** Duplicate atom pair "{}" in SADI restraint line {} of "{}". ***'.format(" ".join(i),
num,
fragment))
pairlist.append(i)
try:
a = atoms[atnames.index(i[0])][2:5]
b = atoms[atnames.index(i[1])][2:5]
except ValueError:
return False
a = [float(x) for x in a]
b = [float(y) for y in b]
dist = atomic_distance(a, b, self.get_cell(fragment))
distances.append(dist)
stdev = std_dev(distances) # Error distribution of
# only do outlier test if standard deviation is suspiciously large:
if stdev > 0.065:
outliers = nalimov_test(distances)
if outliers:
print("\nFragment {}:".format(fragment))
for x in outliers:
pair = ' '.join(pairlist[x])
print('*** Suspicious deviation of atom pair "{}" ({:4.3f} A, median: {:4.3f}) after '
'line {} in {}.txt ***'.format(pair, distances[x], median(distances),
self.get_startline(fragment) + num + 1,
self.get_db_name(fragment))
)
print('*** {} ... ***'.format(restr[num][:60]))
good = False
if (stdev > (2.5 * float(dev))) and good:
print("\nFragment {}:".format(fragment))
print(
'*** Suspicious restraints in SADI line {} with high standard deviation {:4.3f} '
'(median length: {:4.3f} A) ***'.format(num + 1, stdev, median(distances)))
print('*** ' + ' '.join(prefixes + line) + ' ***')
good = False
if good:
return True
else:
return False
