Spaces:
No application file
No application file
# Copyright 2012 by Wibowo Arindrarto. All rights reserved. | |
# This file is part of the Biopython distribution and governed by your | |
# choice of the "Biopython License Agreement" or the "BSD 3-Clause License". | |
# Please see the LICENSE file that should have been included as part of this | |
# package. | |
"""Bio.SearchIO object to model a single database hit.""" | |
from itertools import chain | |
from Bio.SearchIO._utils import allitems, optionalcascade, getattr_str | |
from ._base import _BaseSearchObject | |
from .hsp import HSP | |
class Hit(_BaseSearchObject): | |
"""Class representing a single database hit of a search result. | |
Hit objects are the second-level container in the SearchIO module. They | |
are the objects contained within a QueryResult (see QueryResult). They | |
themselves are container for HSP objects and will contain at least one | |
HSP. | |
To have a quick look at a Hit and its contents, invoke ``print`` on it:: | |
>>> from Bio import SearchIO | |
>>> qresult = next(SearchIO.parse('Blast/mirna.xml', 'blast-xml')) | |
>>> hit = qresult[3] | |
>>> print(hit) | |
Query: 33211 | |
mir_1 | |
Hit: gi|301171322|ref|NR_035857.1| (86) | |
Pan troglodytes microRNA mir-520c (MIR520C), microRNA | |
HSPs: ---- -------- --------- ------ --------------- --------------------- | |
# E-value Bit score Span Query range Hit range | |
---- -------- --------- ------ --------------- --------------------- | |
0 8.9e-20 100.47 60 [1:61] [13:73] | |
1 3.3e-06 55.39 60 [0:60] [13:73] | |
You can invoke ``len`` on a Hit object to see how many HSP objects it contains:: | |
>>> len(hit) | |
2 | |
Hit objects behave very similar to Python lists. You can retrieve the HSP | |
object inside a Hit using the HSP's integer index. Hit objects can also be | |
sliced, which will return a new Hit objects containing only the sliced HSPs:: | |
# HSP items inside the Hit can be retrieved using its integer index | |
>>> hit[0] | |
HSP(hit_id='gi|301171322|ref|NR_035857.1|', query_id='33211', 1 fragments) | |
# slicing returns a new Hit | |
>>> hit | |
Hit(id='gi|301171322|ref|NR_035857.1|', query_id='33211', 2 hsps) | |
>>> hit[:1] | |
Hit(id='gi|301171322|ref|NR_035857.1|', query_id='33211', 1 hsps) | |
>>> print(hit[1:]) | |
Query: 33211 | |
mir_1 | |
Hit: gi|301171322|ref|NR_035857.1| (86) | |
Pan troglodytes microRNA mir-520c (MIR520C), microRNA | |
HSPs: ---- -------- --------- ------ --------------- --------------------- | |
# E-value Bit score Span Query range Hit range | |
---- -------- --------- ------ --------------- --------------------- | |
0 3.3e-06 55.39 60 [0:60] [13:73] | |
Hit objects provide ``filter`` and ``map`` methods, which are analogous to | |
Python's built-in ``filter`` and ``map`` except that they return a new Hit | |
object instead of a list. | |
Here is an example of using ``filter`` to select for HSPs whose e-value is | |
less than 1e-10:: | |
>>> evalue_filter = lambda hsp: hsp.evalue < 1e-10 | |
>>> filtered_hit = hit.filter(evalue_filter) | |
>>> len(hit) | |
2 | |
>>> len(filtered_hit) | |
1 | |
>>> print(filtered_hit) | |
Query: 33211 | |
mir_1 | |
Hit: gi|301171322|ref|NR_035857.1| (86) | |
Pan troglodytes microRNA mir-520c (MIR520C), microRNA | |
HSPs: ---- -------- --------- ------ --------------- --------------------- | |
# E-value Bit score Span Query range Hit range | |
---- -------- --------- ------ --------------- --------------------- | |
0 8.9e-20 100.47 60 [1:61] [13:73] | |
There are also other methods which are counterparts of Python lists' methods | |
with the same names: ``append``, ``index``, ``pop``, and ``sort``. Consult their | |
respective documentations for more details and examples of their usage. | |
""" | |
# attributes we don't want to transfer when creating a new Hit class | |
# from this one | |
_NON_STICKY_ATTRS = ("_items",) | |
def __init__(self, hsps=(), id=None, query_id=None): | |
"""Initialize a Hit object. | |
:param hsps: HSP objects contained in the Hit object | |
:type hsps: iterable yielding HSP | |
:param id: hit ID | |
:type id: string | |
:param query_id: query ID | |
:type query_id: string | |
If multiple HSP objects are used for initialization, they must all | |
have the same ``query_id``, ``query_description``, ``hit_id``, and | |
``hit_description`` properties. | |
""" | |
# default attribute values | |
self._id = id | |
self._id_alt = [] | |
self._query_id = query_id | |
self._description = None | |
self._description_alt = [] | |
self._query_description = None | |
self.attributes = {} | |
self.dbxrefs = [] | |
# TODO - Move this into the for look below in case | |
# hsps is a single use iterator? | |
for attr in ("query_id", "query_description", "hit_id", "hit_description"): | |
# HACK: setting the if clause to '> 1' allows for empty hit objects. | |
# This makes it easier to work with file formats with unpredictable | |
# hit-hsp ordering. The empty hit object itself is nonfunctional, | |
# however, since all its cascading properties are empty. | |
if len({getattr(hsp, attr) for hsp in hsps}) > 1: | |
raise ValueError( | |
"Hit object can not contain HSPs with more than one %s." % attr | |
) | |
self._items = [] | |
for hsp in hsps: | |
# validate each HSP | |
self._validate_hsp(hsp) | |
# and store it them as an instance attribute | |
self.append(hsp) | |
def __repr__(self): | |
"""Return string representation of Hit object.""" | |
return f"Hit(id={self.id!r}, query_id={self.query_id!r}, {len(self)!r} hsps)" | |
def __iter__(self): | |
"""Iterate over hsps.""" | |
return iter(self.hsps) | |
def __len__(self): | |
"""Return number of hsps.""" | |
return len(self.hsps) | |
def __bool__(self): | |
"""Return True if there are hsps.""" | |
return bool(self.hsps) | |
def __contains__(self, hsp): | |
"""Return True if hsp in items.""" | |
return hsp in self._items | |
def __str__(self): | |
"""Return a human readable summary of the Hit object.""" | |
lines = [] | |
# set query id line | |
qid_line = "Query: %s" % self.query_id | |
lines.append(qid_line) | |
if self.query_description: | |
line = " %s" % self.query_description | |
line = line[:77] + "..." if len(line) > 80 else line | |
lines.append(line) | |
# set hit id line | |
hid_line = " Hit: %s" % self.id | |
try: | |
seq_len = self.seq_len | |
except AttributeError: | |
pass | |
else: | |
hid_line += " (%i)" % seq_len | |
lines.append(hid_line) | |
if self.description: | |
line = " %s" % self.description | |
line = line[:77] + "..." if len(line) > 80 else line | |
lines.append(line) | |
# set attributes lines | |
for key, value in sorted(self.attributes.items()): | |
lines.append(f" {key}: {value}") | |
# set dbxrefs line | |
if self.dbxrefs: | |
lines.append("Database cross-references: " + ", ".join(self.dbxrefs)) | |
# set hsp line and table | |
if not self.hsps: | |
lines.append(" HSPs: ?") | |
else: | |
lines.append( | |
" HSPs: %s %s %s %s %s %s" | |
% ("-" * 4, "-" * 8, "-" * 9, "-" * 6, "-" * 15, "-" * 21) | |
) | |
pattern = "%11s %8s %9s %6s %15s %21s" | |
lines.append( | |
pattern | |
% ("#", "E-value", "Bit score", "Span", "Query range", "Hit range") | |
) | |
lines.append( | |
pattern % ("-" * 4, "-" * 8, "-" * 9, "-" * 6, "-" * 15, "-" * 21) | |
) | |
for idx, hsp in enumerate(self.hsps): | |
# evalue | |
evalue = getattr_str(hsp, "evalue", fmt="%.2g") | |
# bitscore | |
bitscore = getattr_str(hsp, "bitscore", fmt="%.2f") | |
# alignment length | |
aln_span = getattr_str(hsp, "aln_span") | |
# query region | |
query_start = getattr_str(hsp, "query_start") | |
query_end = getattr_str(hsp, "query_end") | |
query_range = f"[{query_start}:{query_end}]" | |
# max column length is 18 | |
query_range = ( | |
query_range[:13] + "~]" if len(query_range) > 15 else query_range | |
) | |
# hit region | |
hit_start = getattr_str(hsp, "hit_start") | |
hit_end = getattr_str(hsp, "hit_end") | |
hit_range = f"[{hit_start}:{hit_end}]" | |
hit_range = hit_range[:19] + "~]" if len(hit_range) > 21 else hit_range | |
# append the hsp row | |
lines.append( | |
pattern % (idx, evalue, bitscore, aln_span, query_range, hit_range) | |
) | |
return "\n".join(lines) | |
def __getitem__(self, idx): | |
"""Return the HSP object at the given index.""" | |
# if key is slice, return a new Hit instance | |
if isinstance(idx, slice): | |
obj = self.__class__(self.hsps[idx]) | |
self._transfer_attrs(obj) | |
return obj | |
return self._items[idx] | |
def __setitem__(self, idx, hsps): | |
"""Assign hsps to index idx.""" | |
# handle case if hsps is a list of hsp | |
if isinstance(hsps, (list, tuple)): | |
for hsp in hsps: | |
self._validate_hsp(hsp) | |
else: | |
self._validate_hsp(hsps) | |
self._items[idx] = hsps | |
def __delitem__(self, idx): | |
"""Delete item of index idx.""" | |
del self._items[idx] | |
# hsp properties # | |
def _validate_hsp(self, hsp): | |
"""Validate an HSP object (PRIVATE). | |
Valid HSP objects have the same hit_id as the Hit object ID and the | |
same query_id as the Hit object's query_id. | |
""" | |
if not isinstance(hsp, HSP): | |
raise TypeError("Hit objects can only contain HSP objects.") | |
# HACK: to make validation during __init__ work | |
if self._items: | |
if self.id is not None: | |
if hsp.hit_id != self.id: | |
raise ValueError( | |
"Expected HSP with hit ID %r, found %r instead." | |
% (self.id, hsp.hit_id) | |
) | |
else: | |
self.id = hsp.hit_id | |
if self.description is not None: | |
if hsp.hit_description != self.description: | |
raise ValueError( | |
"Expected HSP with hit description %r, found %r instead." | |
% (self.description, hsp.hit_description) | |
) | |
else: | |
self.description = hsp.hit_description | |
if self.query_id is not None: | |
if hsp.query_id != self.query_id: | |
raise ValueError( | |
"Expected HSP with query ID %r, found %r instead." | |
% (self.query_id, hsp.query_id) | |
) | |
else: | |
self.query_id = hsp.query_id | |
if self.query_description is not None: | |
if hsp.query_description != self.query_description: | |
raise ValueError( | |
"Expected HSP with query description %r, found %r instead." | |
% (self.query_description, hsp.query_description) | |
) | |
else: | |
self.query_description = hsp.query_description | |
# properties # | |
description = optionalcascade( | |
"_description", "hit_description", """Hit description""" | |
) | |
query_description = optionalcascade( | |
"_query_description", | |
"query_description", | |
"""Description of the query that produced the hit""", | |
) | |
id = optionalcascade("_id", "hit_id", """Hit ID string.""") | |
query_id = optionalcascade( | |
"_query_id", "query_id", """ID string of the query that produced the hit""" | |
) | |
# returns all hsps | |
hsps = allitems(doc="""HSP objects contained in the Hit""") | |
def id_all(self): | |
"""Alternative ID(s) of the Hit.""" | |
return [self.id] + self._id_alt | |
def description_all(self): | |
"""Alternative descriptions of the Hit.""" | |
return [self.description] + self._description_alt | |
def fragments(self): | |
"""Access the HSPFragment objects contained in the Hit.""" | |
return list(chain(*self._items)) | |
# public methods # | |
def append(self, hsp): | |
"""Add a HSP object to the end of Hit. | |
Parameters | |
hsp -- HSP object to append. | |
Any HSP object appended must have the same ``hit_id`` property as the | |
Hit object's ``id`` property and the same ``query_id`` property as the | |
Hit object's ``query_id`` property. | |
""" | |
self._validate_hsp(hsp) | |
self._items.append(hsp) | |
def filter(self, func=None): | |
"""Create new Hit object whose HSP objects pass the filter function. | |
:param func: function for filtering | |
:type func: callable, accepts HSP, returns bool | |
``filter`` is analogous to Python's built-in ``filter`` function, except | |
that instead of returning a list it returns a ``Hit`` object. Here is an | |
example of using ``filter`` to select for HSPs having bitscores bigger | |
than 60:: | |
>>> from Bio import SearchIO | |
>>> qresult = next(SearchIO.parse('Blast/mirna.xml', 'blast-xml')) | |
>>> hit = qresult[3] | |
>>> evalue_filter = lambda hsp: hsp.bitscore > 60 | |
>>> filtered_hit = hit.filter(evalue_filter) | |
>>> len(hit) | |
2 | |
>>> len(filtered_hit) | |
1 | |
>>> print(filtered_hit) | |
Query: 33211 | |
mir_1 | |
Hit: gi|301171322|ref|NR_035857.1| (86) | |
Pan troglodytes microRNA mir-520c (MIR520C), microRNA | |
HSPs: ---- -------- --------- ------ --------------- --------------------- | |
# E-value Bit score Span Query range Hit range | |
---- -------- --------- ------ --------------- --------------------- | |
0 8.9e-20 100.47 60 [1:61] [13:73] | |
""" | |
hsps = list(filter(func, self.hsps)) | |
if hsps: | |
obj = self.__class__(hsps) | |
self._transfer_attrs(obj) | |
return obj | |
def index(self, hsp): | |
"""Return the index of a given HSP object, zero-based. | |
:param hsp: object to look up | |
:type hsp: HSP | |
""" | |
return self._items.index(hsp) | |
def map(self, func=None): | |
"""Create new Hit object, mapping the given function to its HSPs. | |
:param func: function for mapping | |
:type func: callable, accepts HSP, returns HSP | |
``map`` is analogous to Python's built-in ``map`` function. It is applied to | |
all HSPs contained in the Hit object and returns a new Hit object. | |
""" | |
if func is not None: | |
hsps = [func(x) for x in self.hsps[:]] # this creates a shallow copy | |
else: | |
hsps = self.hsps[:] | |
if hsps: | |
obj = self.__class__(hsps) | |
self._transfer_attrs(obj) | |
return obj | |
def pop(self, index=-1): | |
"""Remove and returns the HSP object at the specified index. | |
:param index: index of HSP object to pop | |
:type index: int | |
""" | |
return self._items.pop(index) | |
def sort(self, key=None, reverse=False, in_place=True): | |
"""Sort the HSP objects. | |
:param key: sorting function | |
:type key: callable, accepts HSP, returns key for sorting | |
:param reverse: whether to reverse sorting results or no | |
:type reverse: bool | |
:param in_place: whether to do in-place sorting or no | |
:type in_place: bool | |
``sort`` defaults to sorting in-place, to mimic Python's ``list.sort`` | |
method. If you set the ``in_place`` argument to False, it will treat | |
return a new, sorted Hit object and keep the initial one unsorted | |
""" | |
if in_place: | |
self._items.sort(key=key, reverse=reverse) | |
else: | |
hsps = self.hsps[:] | |
hsps.sort(key=key, reverse=reverse) | |
obj = self.__class__(hsps) | |
self._transfer_attrs(obj) | |
return obj | |
# if not used as a module, run the doctest | |
if __name__ == "__main__": | |
from Bio._utils import run_doctest | |
run_doctest() | |