Bio.SearchIO package

Module contents

Biopython interface for sequence search program outputs.

The SearchIO submodule provides parsers, indexers, and writers for outputs from various sequence search programs. It provides an API similar to SeqIO and AlignIO, with the following main functions: parse, read, to_dict, index, index_db, write, and convert.

SearchIO parses a search output file’s contents into a hierarchy of four nested objects: QueryResult, Hit, HSP, and HSPFragment. Each of them models a part of the search output file:

  • QueryResult represents a search query. This is the main object returned by the input functions and it contains all other objects.

  • Hit represents a database hit,

  • HSP represents high-scoring alignment region(s) in the hit,

  • HSPFragment represents a contiguous alignment within the HSP

In addition to the four objects above, SearchIO is also tightly integrated with the SeqRecord objects (see SeqIO) and MultipleSeqAlignment objects (see AlignIO). SeqRecord objects are used to store the actual matching hit and query sequences, while MultipleSeqAlignment objects stores the alignment between them.

A detailed description of these objects’ features and their example usages are available in their respective documentations.

Input

The main function for parsing search output files is Bio.SearchIO.parse(…). This function parses a given search output file and returns a generator object that yields one QueryResult object per iteration.

parse takes two arguments: 1) a file handle or a filename of the input file (the search output file) and 2) the format name.

>>> from Bio import SearchIO
>>> for qresult in SearchIO.parse('Blast/mirna.xml', 'blast-xml'):
...     print("%s %s" % (qresult.id, qresult.description))
...
33211 mir_1
33212 mir_2
33213 mir_3

SearchIO also provides the Bio.SearchIO.read(…) function, which is intended for use on search output files containing only one query. read returns one QueryResult object and will raise an exception if the source file contains more than one queries:

>>> qresult = SearchIO.read('Blast/xml_2226_blastp_004.xml', 'blast-xml')
>>> print("%s %s" % (qresult.id, qresult.description))
...
gi|11464971:4-101 pleckstrin [Mus musculus]
>>> SearchIO.read('Blast/mirna.xml', 'blast-xml')
Traceback (most recent call last):
...
ValueError: ...

For accessing search results of large output files, you may use the indexing functions Bio.SearchIO.index(…) or Bio.SearchIO.index_db(…). They have a similar interface to their counterparts in SeqIO and AlignIO, with the addition of optional, format-specific keyword arguments.

Output

SearchIO has writing support for several formats, accessible from the Bio.SearchIO.write(…) function. This function returns a tuple of four numbers: the number of QueryResult, Hit, HSP, and HSPFragment written:

qresults = SearchIO.parse('Blast/mirna.xml', 'blast-xml')
SearchIO.write(qresults, 'results.tab', 'blast-tab')
<stdout> (3, 239, 277, 277)

Note that different writers may require different attribute values of the SearchIO objects. This limits the scope of writable search results to search results possessing the required attributes.

For example, the writer for HMMER domain table output requires the conditional e-value attribute from each HSP object, among others. If you try to write to the HMMER domain table format and your HSPs do not have this attribute, an exception will be raised.

Conversion

SearchIO provides a shortcut function Bio.SearchIO.convert(…) to convert a given file into another format. Under the hood, convert simply parses a given output file and writes it to another using the parse and write functions.

Note that the same restrictions found in Bio.SearchIO.write(…) applies to the convert function as well.

Conventions

The main goal of creating SearchIO is to have a common, easy to use interface across different search output files. As such, we have also created some conventions / standards for SearchIO that extend beyond the common object model. These conventions apply to all files parsed by SearchIO, regardless of their individual formats.

Python-style sequence coordinates

When storing sequence coordinates (start and end values), SearchIO uses the Python-style slice convention: zero-based and half-open intervals. For example, if in a BLAST XML output file the start and end coordinates of an HSP are 10 and 28, they would become 9 and 28 in SearchIO. The start coordinate becomes 9 because Python indices start from zero, while the end coordinate remains 28 as Python slices omit the last item in an interval.

Beside giving you the benefits of standardization, this convention also makes the coordinates usable for slicing sequences. For example, given a full query sequence and the start and end coordinates of an HSP, one can use the coordinates to extract part of the query sequence that results in the database hit.

When these objects are written to an output file using SearchIO.write(…), the coordinate values are restored to their respective format’s convention. Using the example above, if the HSP would be written to an XML file, the start and end coordinates would become 10 and 28 again.

Sequence coordinate order

Some search output format reverses the start and end coordinate sequences according to the sequence’s strand. For example, in BLAST plain text format if the matching strand lies in the minus orientation, then the start coordinate will always be bigger than the end coordinate.

In SearchIO, start coordinates are always smaller than the end coordinates, regardless of their originating strand. This ensures consistency when using the coordinates to slice full sequences.

Note that this coordinate order convention is only enforced in the HSPFragment level. If an HSP object has several HSPFragment objects, each individual fragment will conform to this convention. But the order of the fragments within the HSP object follows what the search output file uses.

Similar to the coordinate style convention, the start and end coordinates’ order are restored to their respective formats when the objects are written using Bio.SearchIO.write(…).

Frames and strand values

SearchIO only allows -1, 0, 1 and None as strand values. For frames, the only allowed values are integers from -3 to 3 (inclusive) and None. Both of these are standard Biopython conventions.

Supported Formats

Below is a list of search program output formats supported by SearchIO.

Support for parsing, indexing, and writing:

  • blast-tab - BLAST+ tabular output. Both variants without comments

    (-m 6 flag) and with comments (-m 7 flag) are supported.

  • blast-xml - BLAST+ XML output.

  • blat-psl - The default output of BLAT (PSL format). Variants with or

    without header are both supported. PSLX (PSL + sequences) is also supported.

  • hmmer3-tab - HMMER3 table output.

  • hmmer3-domtab - HMMER3 domain table output. When using this format, the

    program name has to be specified. For example, for parsing hmmscan output, the name would be ‘hmmscan-domtab’.

Support for parsing and indexing:

  • exonerate-text - Exonerate plain text output.

  • exonerate-vulgar - Exonerate vulgar line.

  • exonerate-cigar - Exonerate cigar line.

  • fasta-m10 - Bill Pearson’s FASTA -m 10 output.

  • hmmer3-text - HMMER3 regular text output format. Supported HMMER3

    subprograms are hmmscan, hmmsearch, and phmmer.

  • hmmer2-text - HMMER2 regular text output format. Supported HMMER2

    subprograms are hmmpfam, hmmsearch.

Support for parsing:

  • blast-text - BLAST+ plain text output.

  • hhsuite2-text - HHSUITE plain text output.

Each of these formats have different keyword arguments available for use with the main SearchIO functions. More details and examples are available in each of the format’s documentation.

Bio.SearchIO.read(handle, format=None, **kwargs)

Turn a search output file containing one query into a single QueryResult.

  • handle - Handle to the file, or the filename as a string.

  • format - Lower case string denoting one of the supported formats.

  • kwargs - Format-specific keyword arguments.

read is used for parsing search output files containing exactly one query:

>>> from Bio import SearchIO
>>> qresult = SearchIO.read('Blast/xml_2226_blastp_004.xml', 'blast-xml')
>>> print("%s %s" % (qresult.id, qresult.description))
...
gi|11464971:4-101 pleckstrin [Mus musculus]

If the given handle has no results, an exception will be raised:

>>> from Bio import SearchIO
>>> qresult = SearchIO.read('Blast/tab_2226_tblastn_002.txt', 'blast-tab')
Traceback (most recent call last):
...
ValueError: No query results found in handle

Similarly, if the given handle has more than one results, an exception will also be raised:

>>> from Bio import SearchIO
>>> qresult = SearchIO.read('Blast/tab_2226_tblastn_001.txt', 'blast-tab')
Traceback (most recent call last):
...
ValueError: More than one query results found in handle

Like parse, read may also accept keyword argument(s) depending on the search output file format.

Bio.SearchIO.parse(handle, format=None, **kwargs)

Iterate over search tool output file as QueryResult objects.

Arguments:
  • handle - Handle to the file, or the filename as a string.

  • format - Lower case string denoting one of the supported formats.

  • kwargs - Format-specific keyword arguments.

This function is used to iterate over each query in a given search output file:

>>> from Bio import SearchIO
>>> qresults = SearchIO.parse('Blast/mirna.xml', 'blast-xml')
>>> qresults
<generator object ...>
>>> for qresult in qresults:
...     print("Search %s has %i hits" % (qresult.id, len(qresult)))
...
Search 33211 has 100 hits
Search 33212 has 44 hits
Search 33213 has 95 hits

Depending on the file format, parse may also accept additional keyword argument(s) that modifies the behavior of the format parser. Here is a simple example, where the keyword argument enables parsing of a commented BLAST tabular output file:

>>> from Bio import SearchIO
>>> for qresult in SearchIO.parse('Blast/mirna.tab', 'blast-tab', comments=True):
...     print("Search %s has %i hits" % (qresult.id, len(qresult)))
...
Search 33211 has 100 hits
Search 33212 has 44 hits
Search 33213 has 95 hits
Bio.SearchIO.to_dict(qresults, key_function=None)

Turn a QueryResult iterator or list into a dictionary.

  • qresults - Iterable returning QueryResult objects.

  • key_function - Optional callback function which when given a

    QueryResult object should return a unique key for the dictionary. Defaults to using .id of the result.

This function enables access of QueryResult objects from a single search output file using its identifier.

>>> from Bio import SearchIO
>>> qresults = SearchIO.parse('Blast/wnts.xml', 'blast-xml')
>>> search_dict = SearchIO.to_dict(qresults)
>>> list(search_dict)
['gi|195230749:301-1383', 'gi|325053704:108-1166', ..., 'gi|53729353:216-1313']
>>> search_dict['gi|156630997:105-1160']
QueryResult(id='gi|156630997:105-1160', 5 hits)

By default, the dictionary key is the QueryResult’s string ID. This may be changed by supplying a callback function that returns the desired identifier. Here is an example using a function that removes the ‘gi|’ part in the beginning of the QueryResult ID.

>>> from Bio import SearchIO
>>> qresults = SearchIO.parse('Blast/wnts.xml', 'blast-xml')
>>> key_func = lambda qresult: qresult.id.split('|')[1]
>>> search_dict = SearchIO.to_dict(qresults, key_func)
>>> list(search_dict)
['195230749:301-1383', '325053704:108-1166', ..., '53729353:216-1313']
>>> search_dict['156630997:105-1160']
QueryResult(id='gi|156630997:105-1160', 5 hits)

Note that the callback function does not change the QueryResult’s ID value. It only changes the key value used to retrieve the associated QueryResult.

As this function loads all QueryResult objects into memory, it may be unsuitable for dealing with files containing many queries. In that case, it is recommended that you use either index or index_db.

Since Python 3.7, the default dict class maintains key order, meaning this dictionary will reflect the order of records given to it. For CPython, this was already implemented in 3.6.

As of Biopython 1.73, we explicitly use OrderedDict for CPython older than 3.6 (and for other Python older than 3.7) so that you can always assume the record order is preserved.

Bio.SearchIO.index(filename, format=None, key_function=None, **kwargs)

Indexes a search output file and returns a dictionary-like object.

  • filename - string giving name of file to be indexed

  • format - Lower case string denoting one of the supported formats.

  • key_function - Optional callback function which when given a

    QueryResult should return a unique key for the dictionary.

  • kwargs - Format-specific keyword arguments.

Index returns a pseudo-dictionary object with QueryResult objects as its values and a string identifier as its keys. The function is mainly useful for dealing with large search output files, as it enables access to any given QueryResult object much faster than using parse or read.

Index works by storing in-memory the start locations of all queries in a file. When a user requested access to the query, this function will jump to its start position, parse the whole query, and return it as a QueryResult object:

>>> from Bio import SearchIO
>>> search_idx = SearchIO.index('Blast/wnts.xml', 'blast-xml')
>>> search_idx
SearchIO.index('Blast/wnts.xml', 'blast-xml', key_function=None)
>>> sorted(search_idx)
['gi|156630997:105-1160', 'gi|195230749:301-1383', ..., 'gi|53729353:216-1313']
>>> search_idx['gi|195230749:301-1383']
QueryResult(id='gi|195230749:301-1383', 5 hits)
>>> search_idx.close()

If the file is BGZF compressed, this is detected automatically. Ordinary GZIP files are not supported:

>>> from Bio import SearchIO
>>> search_idx = SearchIO.index('Blast/wnts.xml.bgz', 'blast-xml')
>>> search_idx
SearchIO.index('Blast/wnts.xml.bgz', 'blast-xml', key_function=None)
>>> search_idx['gi|195230749:301-1383']
QueryResult(id='gi|195230749:301-1383', 5 hits)
>>> search_idx.close()

You can supply a custom callback function to alter the default identifier string. This function should accept as its input the QueryResult ID string and return a modified version of it.

>>> from Bio import SearchIO
>>> key_func = lambda id: id.split('|')[1]
>>> search_idx = SearchIO.index('Blast/wnts.xml', 'blast-xml', key_func)
>>> search_idx
SearchIO.index('Blast/wnts.xml', 'blast-xml', key_function=<function <lambda> at ...>)
>>> sorted(search_idx)
['156630997:105-1160', ..., '371502086:108-1205', '53729353:216-1313']
>>> search_idx['156630997:105-1160']
QueryResult(id='gi|156630997:105-1160', 5 hits)
>>> search_idx.close()

Note that the callback function does not change the QueryResult’s ID value. It only changes the key value used to retrieve the associated QueryResult.

Bio.SearchIO.index_db(index_filename, filenames=None, format=None, key_function=None, **kwargs)

Indexes several search output files into an SQLite database.

  • index_filename - The SQLite filename.

  • filenames - List of strings specifying file(s) to be indexed, or when

    indexing a single file this can be given as a string. (optional if reloading an existing index, but must match)

  • format - Lower case string denoting one of the supported formats.

    (optional if reloading an existing index, but must match)

  • key_function - Optional callback function which when given a

    QueryResult identifier string should return a unique key for the dictionary.

  • kwargs - Format-specific keyword arguments.

The index_db function is similar to index in that it indexes the start position of all queries from search output files. The main difference is instead of storing these indices in-memory, they are written to disk as an SQLite database file. This allows the indices to persist between Python sessions. This enables access to any queries in the file without any indexing overhead, provided it has been indexed at least once.

>>> from Bio import SearchIO
>>> idx_filename = ":memory:" # Use a real filename, this is in RAM only!
>>> db_idx = SearchIO.index_db(idx_filename, 'Blast/mirna.xml', 'blast-xml')
>>> sorted(db_idx)
['33211', '33212', '33213']
>>> db_idx['33212']
QueryResult(id='33212', 44 hits)
>>> db_idx.close()

index_db can also index multiple files and store them in the same database, making it easier to group multiple search files and access them from a single interface.

>>> from Bio import SearchIO
>>> idx_filename = ":memory:" # Use a real filename, this is in RAM only!
>>> files = ['Blast/mirna.xml', 'Blast/wnts.xml']
>>> db_idx = SearchIO.index_db(idx_filename, files, 'blast-xml')
>>> sorted(db_idx)
['33211', '33212', '33213', 'gi|156630997:105-1160', ..., 'gi|53729353:216-1313']
>>> db_idx['33212']
QueryResult(id='33212', 44 hits)
>>> db_idx.close()

One common example where this is helpful is if you had a large set of query sequences (say ten thousand) which you split into ten query files of one thousand sequences each in order to run as ten separate BLAST jobs on a cluster. You could use index_db to index the ten BLAST output files together for seamless access to all the results as one dictionary.

Note that ‘:memory:’ rather than an index filename tells SQLite to hold the index database in memory. This is useful for quick tests, but using the Bio.SearchIO.index(…) function instead would use less memory.

BGZF compressed files are supported, and detected automatically. Ordinary GZIP compressed files are not supported.

See also Bio.SearchIO.index(), Bio.SearchIO.to_dict(), and the Python module glob which is useful for building lists of files.

Bio.SearchIO.write(qresults, handle, format=None, **kwargs)

Write QueryResult objects to a file in the given format.

  • qresults - An iterator returning QueryResult objects or a single

    QueryResult object.

  • handle - Handle to the file, or the filename as a string.

  • format - Lower case string denoting one of the supported formats.

  • kwargs - Format-specific keyword arguments.

The write function writes QueryResult object(s) into the given output handle / filename. You can supply it with a single QueryResult object or an iterable returning one or more QueryResult objects. In both cases, the function will return a tuple of four values: the number of QueryResult, Hit, HSP, and HSPFragment objects it writes to the output file:

from Bio import SearchIO
qresults = SearchIO.parse('Blast/mirna.xml', 'blast-xml')
SearchIO.write(qresults, 'results.tab', 'blast-tab')
<stdout> (3, 239, 277, 277)

The output of different formats may be adjusted using the format-specific keyword arguments. Here is an example that writes BLAT PSL output file with a header:

from Bio import SearchIO
qresults = SearchIO.parse('Blat/psl_34_001.psl', 'blat-psl')
SearchIO.write(qresults, 'results.tab', 'blat-psl', header=True)
<stdout> (2, 13, 22, 26)
Bio.SearchIO.convert(in_file, in_format, out_file, out_format, in_kwargs=None, out_kwargs=None)

Convert between two search output formats, return number of records.

  • in_file - Handle to the input file, or the filename as string.

  • in_format - Lower case string denoting the format of the input file.

  • out_file - Handle to the output file, or the filename as string.

  • out_format - Lower case string denoting the format of the output file.

  • in_kwargs - Dictionary of keyword arguments for the input function.

  • out_kwargs - Dictionary of keyword arguments for the output function.

The convert function is a shortcut function for parse and write. It has the same return type as write. Format-specific arguments may be passed to the convert function, but only as dictionaries.

Here is an example of using convert to convert from a BLAST+ XML file into a tabular file with comments:

from Bio import SearchIO
in_file = 'Blast/mirna.xml'
in_fmt = 'blast-xml'
out_file = 'results.tab'
out_fmt = 'blast-tab'
out_kwarg = {'comments': True}
SearchIO.convert(in_file, in_fmt, out_file, out_fmt, out_kwargs=out_kwarg)
<stdout> (3, 239, 277, 277)

Given that different search output file provide different statistics and different level of details, the convert function is limited only to converting formats that have the same statistics and for conversion to formats with the same level of detail, or less.

For example, converting from a BLAST+ XML output to a HMMER table file is not possible, as these are two search programs with different kinds of statistics. In theory, you may provide the necessary values required by the HMMER table file (e.g. conditional e-values, envelope coordinates, etc). However, these values are likely to hold little meaning as they are not true HMMER-computed values.

Another example is converting from BLAST+ XML to BLAST+ tabular file. This is possible, as BLAST+ XML provide all the values necessary to create a BLAST+ tabular file. However, the reverse conversion may not be possible. There are more details covered in the XML file that are not found in a tabular file (e.g. the lambda and kappa values)