def _multivec(filepath,
output_file,
assembly,
tile_size,
chromsizes_filename,
starting_resolution,
row_infos_filename=None):
'''
Aggregate a multivec file.
This is a file containing nxn data that is aggregated along only one axis.
This data should be in an HDF5 file where each dataset is named for a
chromosome and contains a 'resolutions' group containing values for the
base level resolution.
Example: f['chr1']['reslutions']['1000'] = [[1,2,3],[4,5,6]]
The resulting data will be organized by resolution and chromosome.
Example: f_out['chr1']['resolutions']['5000']=[[1000,2000,3000],[4000,5000,6000]]
Aggregation is currently done by summing adjacent values.
'''
f_in = h5py.File(filepath, 'r')
if output_file is None:
output_file = op.splitext(filepath)[0] + ".multires.mv5"
(chrom_info, chrom_names,
chrom_sizes) = cch.load_chromsizes(chromsizes_filename, assembly)
if method == 'maxtotal':
pass
if method == 'logsumexp':
def agg(x):
a = x.T.reshape((x.shape[1], -1, 2))
return sm.logsumexp(a, axis=2).T
else:
agg = lambda x: x.T.reshape((x.shape[1], -1, 2)).sum(axis=2).T
print("agg:", agg)
if row_infos_filename is not None:
with open(row_infos_filename, 'r') as fr:
row_infos = [l.strip().encode('utf8') for l in fr]
else:
row_infos = None
print("row_infos:", row_infos)
cmv.create_multivec_multires(
f_in,
chromsizes=zip(chrom_names, chrom_sizes),
agg=lambda x: np.nansum(x.T.reshape((x.shape[1], -1, 2)), axis=2).T,
starting_resolution=starting_resolution,
tile_size=tile_size,
output_file=output_file,
row_infos=row_infos)
def _bedgraph_to_multivec(filepaths, output_file, assembly, chrom_col,
from_pos_col, to_pos_col, value_col, has_header,
chunk_size, nan_value, chromsizes_filename,
starting_resolution, num_rows, format,
row_infos_filename, tile_size, method):
print('chrom_col:', chrom_col)
with tempfile.TemporaryDirectory() as td:
print('temporary dir:', td)
temp_file = op.join(td, 'temp.mv5')
f_out = h5py.File(temp_file, 'w')
(chrom_info, chrom_names,
chrom_sizes) = cch.load_chromsizes(chromsizes_filename, assembly)
if row_infos_filename is not None:
with open(row_infos_filename, 'r') as fr:
row_infos = [l.strip().encode('utf8') for l in fr]
else:
row_infos = None
for chrom in chrom_info.chrom_order:
f_out.create_dataset(chrom, (math.ceil(
chrom_info.chrom_lengths[chrom] / starting_resolution),
num_rows * len(filepaths)),
fillvalue=np.nan,
compression='gzip')
def bedline_to_chrom_start_end_vector(bedlines, row_infos=None):
chrom_set = set()
start_set = set()
end_set = set()
all_vector = []
for bedline in bedlines:
parts = bedline.strip().split()
chrom = parts[chrom_col - 1]
start = int(parts[from_pos_col - 1])
end = int(parts[to_pos_col - 1])
vector = [
float(f) if not f == 'NA' else np.nan
for f in parts[value_col - 1:value_col - 1 + num_rows]
]
chrom_set.add(chrom)
start_set.add(start)
end_set.add(end)
if len(chrom_set) > 1:
raise ValueError("Chromosomes don't match in these lines:",
bedlines)
if len(start_set) > 1:
raise ValueError(
"Start positions don't match in these lines:",
bedlines)
if len(end_set) > 1:
raise ValueError(
"End positions don't match in these lines:", bedlines)
all_vector += vector
return (list(chrom_set)[0], list(start_set)[0], list(end_set)[0],
all_vector)
if format == 'epilogos':
cmv.bedfile_to_multivec(filepaths, f_out,
epilogos_bedline_to_vector,
starting_resolution, has_header,
chunk_size)
elif format == 'states':
assert (
row_infos != None
), "A row_infos file must be provided for --format = 'states' "
states_dic = {row_infos[x]: x for x in range(len(row_infos))}
cmv.bedfile_to_multivec(filepaths, f_out, states_bedline_to_vector,
starting_resolution, has_header,
chunk_size, states_dic)
else:
cmv.bedfile_to_multivec(filepaths, f_out,
bedline_to_chrom_start_end_vector,
starting_resolution, has_header,
chunk_size)
f_out.close()
tf = temp_file
f_in = h5py.File(tf, 'r')
if output_file is None:
output_file = op.splitext(filepaths[0])[0] + '.multires.mv5'
print('output_file:', output_file)
# Override the output file if it existts
if op.exists(output_file):
os.remove(output_file)
if method == 'logsumexp':
def agg(x):
# newshape = (x.shape[2], -1, 2)
# b = x.T.reshape((-1,))
a = x.T.reshape((x.shape[1], -1, 2))
# this is going to be an odd way to get rid of nan
# values
orig_shape = a.shape
na = a.reshape((-1, ))
SMALL_NUM = -1e8
NAN_THRESHOLD_NUM = SMALL_NUM / 100
if np.nanmin(na) < NAN_THRESHOLD_NUM:
raise ValueError(
"Error removing nan's when running logsumexp aggregation"
)
na[np.isnan(na)] = SMALL_NUM
na = na.reshape(orig_shape)
res = sm.logsumexp(a, axis=2).T
nres = res.reshape((-1, ))
# print("nres:", np.nansum(nres < NAN_THRESHOLD_NUM))
nres[nres < NAN_THRESHOLD_NUM] = np.nan
res = nres.reshape(res.shape)
# print("res:", np.nansum(res.reshape((-1,))))
return res
else:
agg = lambda x: x.T.reshape((x.shape[1], -1, 2)).sum(axis=2).T
cmv.create_multivec_multires(f_in,
chromsizes=zip(chrom_names, chrom_sizes),
agg=agg,
starting_resolution=starting_resolution,
tile_size=tile_size,
output_file=output_file,
row_infos=row_infos)
def bigwigs_to_multivec(
filepaths,
output_file,
assembly,
chromsizes_filename,
row_infos_filename,
tile_size,
):
with tempfile.TemporaryDirectory() as td:
print("temporary dir:", td)
temp_file = op.join(td, "temp.mv5")
f_out = h5py.File(temp_file, "w")
(chrom_info, chrom_names,
chrom_lengths) = cch.load_chromsizes(chromsizes_filename, assembly)
if row_infos_filename is not None:
with open(row_infos_filename, "r") as f:
row_infos = [line.strip().encode("utf8") for line in f]
else:
row_infos = None
starting_resolution = 1
resolution = starting_resolution
for chrom in chrom_info.chrom_order:
f_out.create_dataset(
chrom,
(
math.ceil(
chrom_info.chrom_lengths[chrom] / starting_resolution),
len(filepaths),
),
fillvalue=np.nan,
compression="gzip",
)
# Fill in data for each bigwig file.
for bw_index, bw_file in tqdm(list(enumerate(filepaths)),
desc="bigwigs"):
if bbi.is_bigwig(bw_file):
chromsizes = bbi.chromsizes(bw_file)
matching_chromosomes = set(chromsizes.keys()).intersection(
set(chrom_names))
# Fill in data for each resolution of a bigwig file.
for chr_name in matching_chromosomes:
print("chr_name:", chr_name, resolution)
chr_len = chrom_info.chrom_lengths[chr_name]
chr_shape = (math.ceil(chr_len / resolution),
len(filepaths))
arr = bbi.fetch(bw_file,
chr_name,
0,
chr_len,
chr_shape[0],
summary="sum")
f_out[chr_name][:, bw_index] = arr
else:
print(f"{bw_file} not is_bigwig")
f_out.flush()
f_out.close()
tf = temp_file
f_in = h5py.File(tf, "r")
def agg(x):
return x.T.reshape((x.shape[1], -1, 2)).sum(axis=2).T
cmv.create_multivec_multires(
f_in,
chromsizes=zip(chrom_names, chrom_lengths),
agg=agg,
starting_resolution=starting_resolution,
tile_size=tile_size,
output_file=output_file,
row_infos=row_infos,
)
