def __getitem__(self, idx):
if self.fasta_extractors is None:
self.fasta_extractors = FastaStringExtractor(
self.fasta_file,
use_strand=False, # self.use_strand,
force_upper=self.force_upper)
interval, labels = self.bed[idx]
if self.auto_resize_len:
# automatically resize the sequence to cerat
interval = resize_interval(interval,
self.auto_resize_len,
anchor='center')
# QUESTION: @kromme - why to we need max_seq_len?
# if self.max_seq_len is not None:
# assert interval.stop - interval.start <= self.max_seq_len
# Run the fasta extractor and transform if necessary
seq = self.fasta_extractors.extract(interval)
return {
"inputs": np.array(seq),
"targets": labels,
"metadata": {
"ranges":
GenomicRanges(interval.chrom, interval.start, interval.stop,
str(idx))
}
}
def test_resize_interval(anchor, ilen):
import pybedtools
dummy_start, dummy_end = 10, 20
dummy_center = int((dummy_start + dummy_end) / 2)
dummy_inter = pybedtools.create_interval_from_list(
['chr2', dummy_start, dummy_end, 'intname'])
ret_inter = resize_interval(dummy_inter, ilen, anchor)
# the original interval was left intact
assert dummy_inter.chrom == 'chr2'
assert dummy_inter.start == dummy_start
assert dummy_inter.end == dummy_end
assert dummy_inter.name == 'intname'
# metadata kept
assert ret_inter.chrom == dummy_inter.chrom
assert ret_inter.name == 'intname'
# desired output width
assert ret_inter.length == ilen
# correct anchor point
if anchor == "start":
assert ret_inter.start == dummy_start
elif anchor == "end":
assert ret_inter.end == dummy_end
elif anchor == "center":
assert int((ret_inter.start + ret_inter.end) / 2) == dummy_center
def extract(self, interval):
"""
Extract the coverage corresponding to the interval from the bbi_files.
returns:
np.array of shape (window,
number of files per annotation,
number of annotation)
"""
if not self.sampling_mode:
assert self.window <= abs(interval.stop - interval.start),\
"""The target window must be smaller than the input length"""
interval = resize_interval(interval,
self.window,
anchor='center')
seq = list()
for bbi_file in self.bbi_files:
bw = pyBigWig.open(bbi_file)
array = bw.values(interval.chrom,
interval.start,
interval.stop, numpy=True)
array[np.isnan(array)] = 0
seq.append(self.norm_dico[bbi_file](array))
seq = np.array(seq).T
if self.sampling_mode:
assert abs(interval.stop - interval.start) % self.window == 0,\
"""Window must divide the input length to use downsampling"""
sampling_length = abs(interval.stop - interval.start) // self.window
if self.sampling_mode == 'downsampling':
seq = seq[::sampling_length]
elif self.sampling_mode == 'mean':
seq = self._calculate_rolling_mean(seq)
else:
raise NameError('sampling_mode must be None, "mean" or "downsampling"')
if self.nb_annotation_type:
nb_files_per_ann = len(self.bbi_files) // self.nb_annotation_type
return seq.reshape((self.window,
nb_files_per_ann,
self.nb_annotation_type))
else:
return seq
def __getitem__(self, idx):
self.fasta_extractor = FastaStringExtractor(self.fasta_file)
# get the intervals
interval, targets = self.bt[idx]
# resize to 500bp
interval = resize_interval(interval, 500, anchor='center')
# extract the sequence
seq = self.fasta_extractor.extract(interval)
# one-hot encode the sequence
seq_onehot = self.transform(seq)
ranges = GenomicRanges.from_interval(interval)
return {"inputs": [seq_onehot], "metadata": [ranges]}
def __call__(self, interval):
return F.resize_interval(interval, self.width, self.anchor)
