HTSeq 2.1.2__cp310-cp310-macosx_10_15_x86_64.whl

HTSeq/StretchVector.py

@@ -0,0 +1,491 @@
+from collections import namedtuple
+import numpy as np
+import warnings
+
+
+Interval = namedtuple('Interval', ['start', 'end'])
+
+
+class StretchVector:
+    """Sparse representation for 'island' of dense data on a long line.
+
+    This class is the most basic object representing 'windows of data' in a sea
+    of unknowns of a linear structure, e.g. a chromosome. The basic design is to
+    store a list of (start, end) intervals, each of them associated with a
+    "stretch", i.e. a dense numpy array.
+
+    In its simplest form, a StretchVector can be thought of as an array with
+    an arbitratily large offset. This can be useful e.g. to plot coverage or
+    signal enrichments while keeping real genomic coordinates.
+
+    The class does more heavy lifting, however, when several separate arrays
+    are used, e.g. to represent "peaks" of signals (ATAC-Seq, ChIP-Seq, etc.).
+    In that situation, the bookkeeping involved in manipulating all offsets
+    correctly can become burdensome quickly, so StretchVector takes care of it.
+    In addition, extension of data to flanks is easy, so if you want to expand
+    an existing data window by 1kb on each side, that can be done directly:
+
+    >>> sv = HTSeq.StretchVector(typecode="d")
+    >>> sv[6789: 8900] = 56.8
+    >>> sv[6000: 7000] = 20 # <-- left extension
+    >>> sv[8000: 9000] = 10 # <-- right extension
+
+    You can use StretchVector as a storage option for higher level objects such
+    as ChromVector and GenomicArray. Those classes support strandedness, unlike
+    StretchVector itself.
+    """
+    _typecode_dict = {
+        "d": np.float32,
+        "i": np.int32,
+        "l": np.int64,
+        "O": object,
+    }
+
+    def __init__(self, typecode):
+        """Create an empty StretchVector of a given type
+
+        Args:
+            typecode ("d", "i", "l", or "O"): The dtype of the stored data. Can
+              be "d" (double, i.e. np.float32), "i" (np.int32), "l" (np.int64),
+              or "O" (generic object). Note that np.float64 data will be recast
+              to np.float32, which might lead to a loss of machine precision.
+
+        Returns:
+            A StretchVector instance of the chosen type.
+        """
+        self.typecode = typecode
+        self.ivs = []
+        self.stretches = []
+
+    def _in_stretch(self, index):
+        if len(self.stretches) == 0:
+            return -1
+
+        if isinstance(index, int):
+            if (index < self.ivs[0].start) or (index >= self.ivs[-1].end):
+                return -1
+            for i, iv in enumerate(self.ivs):
+                if index < iv.start:
+                    return -1
+                if index < iv.end:
+                    return i
+
+    def _get_interval(self, start, end):
+        if len(self.stretches) == 0:
+            return self
+
+        ivs = []
+        stretches = []
+        for i, iv in enumerate(self.ivs):
+            # They end before the start, skip
+            if iv.end <= start:
+                continue
+
+            # They start after the end, skip all remaining
+            if iv.start >= end:
+                break
+
+            # This interval overlap with start-end
+            if (iv.start <= start) and (iv.end <= end):
+                new_iv = Interval(start, iv.end)
+                new_stretch = self.stretches[i][start - iv.start:]
+                ivs.append(new_iv)
+                stretches.append(new_stretch)
+                continue
+
+            if (iv.start <= start) and (iv.end > end):
+                new_iv = Interval(start, end)
+                new_stretch = self.stretches[i][start - iv.start:-(iv.end - end)]
+                ivs.append(new_iv)
+                stretches.append(new_stretch)
+                break
+
+            if (iv.start > start) and (iv.end <= end):
+                new_iv = Interval(iv.start, iv.end)
+                new_stretch = self.stretches[i]
+                ivs.append(new_iv)
+                stretches.append(new_stretch)
+                continue
+
+            if (iv.start > start) and (iv.end > end):
+                new_iv = Interval(iv.start, end)
+                new_stretch = self.stretches[i][:-(iv.end - end)]
+                ivs.append(new_iv)
+                stretches.append(new_stretch)
+                break
+
+        new_cls = self.__class__(self.typecode)
+        new_cls.ivs = ivs
+        new_cls.stretches = stretches
+        return new_cls
+
+    def _set_interval(self, start, end, values):
+        if len(self.stretches) == 0:
+            self.ivs.append(Interval(start, end))
+            self.stretches.append(
+                np.zeros(end - start, self._typecode_dict[self.typecode])
+            )
+            self.stretches[-1][:] = values
+            return len(self.ivs) - 1
+
+        # For each end, there are two possibilities, inside or outside an
+        # existing stretch
+        idx_start = self._in_stretch(start)
+        idx_end = self._in_stretch(end - 1)
+
+        # Neither is in, make a new stretch and delete existing stretches
+        if idx_start == idx_end == -1:
+            new_iv = Interval(start, end)
+            new_stretch = np.zeros(end - start, self._typecode_dict[self.typecode])
+            new_stretch[:] = values
+            new_ivs = []
+            new_stretches = []
+            new_added = False
+            for i, iv in enumerate(self.ivs):
+                # Stretches before
+                if start >= iv.end:
+                    new_ivs.append(iv)
+                    new_stretches.append(self.stretches[i])
+                    continue
+
+                # Add new stretch
+                if not new_added:
+                    new_ivs.append(new_iv)
+                    new_stretches.append(new_stretch)
+
+                # Skip overlapping stretches
+                if (start <= iv.start) and (end >= iv.end):
+                    continue
+
+                # Stretches after
+                new_ivs.append(iv)
+                new_stretches.append(self.stretches[i])
+
+            # Add new stretch if still missing
+            if not new_added:
+                new_ivs.append(new_iv)
+                new_stretches.append(new_stretch)
+
+        # Start is in a stretch, end is not
+        elif (idx_start != -1) and (idx_end == -1):
+            new_iv = Interval(self.ivs[idx_start].start, end)
+            new_stretch = np.zeros(
+                end - self.ivs[idx_start].start,
+                self._typecode_dict[self.typecode],
+            )
+            l1 = start - self.ivs[idx_start].start
+            new_stretch[:l1] = self.stretches[idx_start][:l1]
+            new_stretch[l1:] = values
+
+            new_ivs = self.ivs[:idx_start] + [new_iv]
+            new_stretches = self.stretches[:idx_start] = [new_stretch]
+
+            for i, iv in enumerate(self.ivs[idx_start:], idx_start):
+                # Skip the first one
+                if i == idx_start:
+                    continue
+
+                # Skip overlapping stretches
+                if iv.end < end:
+                    continue
+
+                # Stretches after
+                new_ivs.append(iv)
+                new_stretches.append(self.stretches[i])
+
+        # Start is not in a stretch, end is
+        elif (idx_start == -1) and (idx_end != -1):
+            new_iv = Interval(start, self.ivs[idx_end].end)
+            new_stretch = np.zeros(
+                self.ivs[idx_end].end - start,
+                self._typecode_dict[self.typecode],
+            )
+            l2 = self.ivs[idx_end].end - end
+            if l2 == 0:
+                l2 = -len(self.stretches[idx_end])
+            new_stretch[:-l2] = values
+            if -l2 != len(self.stretches[idx_end]):
+                new_stretch[-l2:] = self.stretches[idx_end][-l2:]
+
+            new_ivs = []
+            new_stretches = []
+            for i, iv in enumerate(self.ivs):
+                # Stretches before
+                if start >= iv.end:
+                    new_ivs.append(iv)
+                    new_stretches.append(self.stretches[i])
+                    continue
+                break
+
+            # New stretch
+            new_ivs.append(new_iv)
+            new_stretches.append(new_stretch)
+
+            # If there are stretches left, add them
+            if idx_end != len(self.ivs) - 1:
+                new_ivs.extend(self.ivs[idx_end+1:])
+                new_stretches.extend(self.stretches[idx_end+1:])
+
+        # Both start and end are in the same stretch
+        elif idx_start == idx_end:
+            l1 = start - self.ivs[idx_start].start
+            l2 = self.ivs[idx_end].end - end
+            if l2 == 0:
+                l2 = -len(self.stretches[idx_end])
+            new_ivs = self.ivs
+            new_stretches = self.stretches
+            new_stretches[idx_start][l1:-l2] = values
+
+        # They are in different stretches
+        else:
+            new_iv = Interval(
+                    self.ivs[idx_start].start,
+                    self.ivs[idx_end].end,
+                )
+            new_stretch = np.zeros(
+                self.ivs[idx_end].end - self.ivs[idx_start].start,
+                self._typecode_dict[self.typecode],
+            )
+            l1 = start - self.ivs[idx_start].start
+            l2 = self.ivs[idx_end].end - end
+            if l2 == 0:
+                l2 = -len(self.stretches[idx_end])
+            new_stretch[:l1] = self.stretches[idx_start][:l1]
+            new_stretch[l1:-l2] = values
+            if -l2 != len(self.stretches[idx_end]):
+                new_stretch[-l2:] = self.stretches[idx_end][-l2:]
+
+            new_ivs = self.ivs[:idx_start] + [iv]
+            if idx_end != len(self.ivs) - 1:
+                new_ivs.extend(self.ivs[idx_end+1:])
+                new_stretches.extend(self.stretches[idx_end+1:])
+
+        self.ivs = new_ivs
+        self.stretches = new_stretches
+
+    def _add_stretch(self, start, end, i_start=0):
+        for i, iv in enumerate(self.ivs, i_start):
+            if start < iv.start:
+                self.ivs.insert(
+                    i,
+                    Interval(start, end),
+                )
+                self.stretches.insert(
+                    i,
+                    np.zeros(end - start, self._typecode_dict[self.typecode])
+                )
+                return i
+
+        self.ivs.append(
+            Interval(start, end),
+        )
+        self.stretches.append(
+            np.zeros(end - start, self._typecode_dict[self.typecode])
+        )
+        return len(self.ivs) - 1
+
+    def __getitem__(self, index):
+        """Get a view of a portion of the StretchVector
+
+
+        Args:
+            index (int, slice, or GenomicInterval): Coordinate or interval to
+              extract. For slices, the stretches from this intervals are
+              *views* of the original array, so changes in them will be
+              reflected in the parent StretchVector as well.
+        Returns:
+            A number of index is an int, containing the value at that site.
+            A StretchVector with adapted stretches if index is anything else.
+
+        """
+        from HTSeq import GenomicInterval
+
+        if isinstance(index, int):
+            idx_iv = self._in_stretch(index)
+            if idx_iv == -1:
+                return None
+            return self.stretches[idx_iv][index - self.ivs[idx_iv].start]
+
+        elif isinstance(index, slice):
+            if index.step is not None and index.step != 1:
+                raise ValueError(
+                        "Striding slices (i.e., step != 1) are not supported")
+            if index.start is None:
+                index.start = 0
+            if index.stop is None:
+                if len(self.ivs) == 0:
+                    raise IndexError('No stretches, cannot find end')
+                index.stop = self.ivs[-1].end
+
+            return self._get_interval(index.start, index.stop)
+
+        elif isinstance(index, GenomicInterval):
+            return self.__getitem__(slice(index.start, index.end))
+
+    def __setitem__(self, index, values):
+        """Set/reset values within or outside the stretch
+
+        Args:
+            index (int, slice, or GenomicInterval): Coordinate or interval to
+              set. These can be within current stretches, between them, outside,
+              or partially overlapping.
+            values (numpy.ndarray or convertible sequence): Values to set or
+              reset at those locations. This will be recast as a numpy array
+              of the appropriate dtype within this function.
+
+        Returns: None
+        """
+        from HTSeq import GenomicInterval
+
+        # Leave dtype out for now for speed, it will be taken care of later on
+        values = np.asarray(values)
+
+        if isinstance(index, int):
+            idx_iv = self._in_stretch(index)
+            if idx_iv == -1:
+                idx_iv = self._add_stretch(index, index + 1)
+            self.stretches[idx_iv][index - self.ivs[idx_iv].start] = values
+            return
+
+        elif isinstance(index, slice):
+            if index.step is not None and index.step != 1:
+                raise ValueError(
+                        "Striding slices (i.e., step != 1) are not supported")
+            if index.start is None:
+                index.start = 0
+            if index.stop is None:
+                if len(self.ivs) == 0:
+                    raise IndexError('No stretches, cannot find end')
+                index.stop = self.ivs[-1].end
+
+            self._set_interval(index.start, index.stop, values)
+
+        elif isinstance(index, GenomicInterval):
+            return self.__setitem__(slice(index.start, index.end), values)
+
+
+    def todense(self):
+        """Dense numpy array of the whole stretch, using NaNs for missing data"""
+        if len(self.ivs) == 0:
+            return np.empty(0, self._typecode_dict[self.typecode])
+
+        if len(self.ivs) == 1:
+            return self.stretches[0].copy()
+
+        # At least two stretches, have to stitch them
+        start = self.ivs[0].start
+        res = np.empty(
+            self.ivs[-1].end - start,
+            self.stretches[0].dtype,
+        )
+        res[:] = np.nan
+
+        for i, iv in enumerate(self.ivs):
+            res[iv.start - start: iv.end - start] = self.stretches[i]
+
+        return res
+
+    @classmethod
+    def from_dense(cls, array, offset=0):
+        """Create from dense array with NaNs
+
+        Args:
+            array (numpy.array): dense array containing NaNs at positions to
+              be skipped.
+            offset (int): Start of the initial interval.
+        """
+        if array.dtype == np.float64:
+            warnings.warn("np.float64 array converted to np.float32")
+            array = array.astype(np.float32)
+
+        for typecode, dtype in cls._typecode_dict.items():
+            if dtype == array.dtype:
+                sv = cls(typecode=typecode)
+                break
+        else:
+            raise TypeError('Typecode not found for dtype: '+str(array.dtype))
+
+        if len(array) == 0:
+            return sv
+
+        flips = np.diff(np.isnan(array)).nonzero()[0]
+
+        # No flips: either all good or all skip
+        if flips.sum() == 0:
+            if np.isnan(array[0]):
+                return sv
+            sv.ivs.append(Interval(offset, offset + len(array)))
+            sv.stretches.append(array.copy())
+            return sv
+
+        # If we start with nan, just increase the offset
+        if np.isnan(array[0]):
+            add_offset = flips[0] + 1
+            offset += add_offset
+            array = array[add_offset:]
+
+            # Single flip means keep the whole rest
+            if len(flips) > 1:
+                flips = flips[1:]
+                flips -= add_offset
+            else:
+                sv.ivs.append(Interval(offset, offset + len(array)))
+                sv.stretches.append(array.copy())
+                return sv
+
+        # Now we have at least one flip left, and we start with a number/object
+        # If we have an odd number of flips, we can forget the last block
+        if len(flips) % 2:
+            end = flips[-1] + 1
+            array = array[:end]
+            flips = flips[:-1]
+
+        # No flip left, all good
+        if len(flips) == 0:
+            sv.ivs.append(Interval(offset, offset + len(array)))
+            sv.stretches.append(array.copy())
+            return sv
+
+        # Now we start and end with a number/object, and there are at least
+        # two flips. Initial stretch
+        sv.ivs.append(Interval(offset, offset + flips[0] + 1))
+        sv.stretches.append(
+                array[:flips[0] + 1]
+        )
+        # Intermediate stretches
+        for i in range((len(flips) // 2) - 1):
+            new_iv = Interval(offset + flips[i * 2 + 1] + 1, offset + flips[i * 2 + 2] + 1)
+            new_stretch = array[flips[i * 2 + 1] + 1: flips[i * 2 + 2] + 1]
+            sv.ivs.append(new_iv)
+            sv.stretches.append(new_stretch)
+        # Final stretch
+        new_iv = Interval(offset + flips[-1] + 1, offset + len(array))
+        new_stretch = array[flips[-1] + 1:]
+        sv.ivs.append(new_iv)
+        sv.stretches.append(new_stretch)
+
+        return sv
+
+    def __iter__(self):
+        """Iterate over intervals and stretches ("islands")."""
+        return zip(self.ivs, self.stretches)
+
+    def copy(self):
+        """Make a copy the StretchVector and of all its stretches"""
+        sv_new = StretchVector(typecode=self.typecode)
+        for iv, stretch in self:
+            sv_new.ivs.append(Interval(iv.start, iv.end))
+            sv_new.stretches.append(stretch.copy())
+        return sv_new
+
+    def shift(self, offset):
+        """Shift all stretch intervals by a constant number
+
+        Args:
+            offset (int): Shift the start and end coordinates of each stretch
+              by this amount.
+        Returns:
+            None. The function acts in place.
+        """
+        for i, iv in self.ivs:
+            self.ivs[i] = Interval(iv.start + offset, iv.end + offset)

HTSeq/_HTSeq_internal.py

@@ -0,0 +1,85 @@
+# Internal HTSeq functions, not part of the API
+import HTSeq
+import numpy
+
+
+def GenomicInterval_range(gi, step):
+    for pos in range(gi.start, gi.end, step):
+        yield HTSeq.GenomicPosition(gi.chrom, pos, gi.strand)
+
+
+def GenomicInterval_xranged(gi, step):
+    if gi.strand == "-":
+        step *= -1
+    for pos in range(gi.start_d, gi.end_d, step):
+        yield HTSeq.GenomicPosition(gi.chrom, pos, gi.strand)
+
+
+def ChromVector_steps(cv):
+    '''Steps over a ChromVector
+
+
+    NOTE: ChromVectors use an offset which is also iv.start compared to their
+          storage objects that start at 0.
+    '''
+    # "Steps" of an ndarray (or memmap?)-storaged ChromVector
+    if isinstance(cv.array, numpy.ndarray):
+        start = cv.iv.start
+        prev_val = None
+        for i in range(cv.iv.start, cv.iv.end):
+            val = cv.array[i - cv.offset]
+            if prev_val is None or val != prev_val:
+                if prev_val is not None:
+                    yield (HTSeq.GenomicInterval(cv.iv.chrom, start, i, cv.iv.strand), prev_val)
+                prev_val = val
+                start = i
+        yield (HTSeq.GenomicInterval(
+            cv.iv.chrom, start, cv.iv.end, cv.iv.strand), prev_val,
+            )
+
+    # Steps of a StepVector-storaged ChromVector
+    elif isinstance(cv.array, HTSeq.StepVector.StepVector):
+        for start, stop, value in cv.array[
+                cv.iv.start - cv.offset: cv.iv.end - cv.offset].get_steps():
+            yield (HTSeq.GenomicInterval(
+                cv.iv.chrom, start + cv.offset, stop + cv.offset, cv.iv.strand), value,
+                )
+
+    # Steps in a StretchVector behave similar to a full numpy array, but uses
+    # np.nan for None and treats each stretch as independent, of course
+    # NOTE: one could optimize this by using np.diff and flips, as we have done
+    # in the StretchVector methods. For now, we leave it like this because
+    # the whole point of StretchVector is to be used for stretches, not steps.
+    elif isinstance(cv.array, HTSeq.StretchVector):
+        for iv, stretch in cv.array:
+            start = cv.offset + iv.start
+            prev_val = None
+            for i, val in enumerate(stretch, cv.offset + iv.start):
+                # Subsequent NaNs, ignore
+                if (prev_val is not None) and numpy.isnan(prev_val) and numpy.isnan(val):
+                    continue
+                if prev_val is None or val != prev_val:
+                    # Delay yield of the first item until you meet the first
+                    # unequal item, i.e. until you know the end of the step
+                    if prev_val is not None:
+                        yield (HTSeq.GenomicInterval(
+                                    cv.iv.chrom,
+                                    start,
+                                    i,
+                                    cv.iv.strand),
+                                prev_val)
+                    prev_val = val
+                    start = i
+            yield (HTSeq.GenomicInterval(
+                cv.iv.chrom, start, cv.offset + iv.end, cv.iv.strand), prev_val,
+                )
+    else:
+        raise SystemError("Unknown array type.")
+
+
+def GenomicArray_steps(ga):
+    """Steps of a GenomicArray are just the chained steps of each ChromVector"""
+    for chrom, chromstrand_dict in ga.chrom_vectors.items():
+        for strand, chrom_vector in chromstrand_dict.items():
+            for iv, val in chrom_vector.steps():
+                yield iv, val