chemfp.search module

Search a FingerprintArena and work with the search results

This module implements the different ways to search a FingerprintArena. The search functions are:

Count the number of hits:

Find all hits at or above a given threshold, sorted arbitrarily:

Find the k-nearest hits at or above a given threshold, sorted by decreasing similarity:

The threshold and k-nearest search results use a SearchResult when a fingerprint is used as a query, or a SearchResults when an arena is used as a query. These internally use a compressed sparse row format.

class chemfp.search.MatrixType

Bases: enum.IntEnum

An enumeration.

NO_DIAGONAL = 2
NxM = 0
NxN = 1
STRICTLY_UPPER_TRIANGULAR = 4
UPPER_TRIANGULAR = 3
from_npz_name()

Return the MatrixType enum value given a chemfp npz type name

Raise KeyError if it does not exist.

get_npz_name(default=None)

Return the chemfp npz type name given a MatrixType enum value

Raise KeyError if it does not exist.

get_npz_names()

Return the list of supported chemfp npz type names

class chemfp.search.SearchResult(search_results, row)

Bases: object

Search results for a query fingerprint against a target arena.

The results contains a list of hits. Hits contain a target index, score, and optional target ids. The hits can be reordered based on score or index.

as_buffer()

Return a Python buffer object for the underlying indices and scores.

This provides a byte-oriented view of the raw data. You probably want to use as_ctypes() or as_numpy_array() to get the indices and scores in a more structured form.

Warning

Do not attempt to access the buffer contents after the search result has been deallocated as that will likely cause a segmentation fault or other severe failure.

Returns:a Python buffer object
as_ctypes()

Return a ctypes view of the underlying indices and scores

Each (index, score) pair is represented as a ctypes structure named Hit with fields index (c_int) and score (c_double).

For example, to get the score of the 5th entry use:

result.as_ctypes()[4].score

This method returns an array of type (Hit*len(search_result)). Modifications to this view will change chemfp’s data values and vice versa. USE WITH CARE!

Warning

Do not attempt to access the ctype array contents after the search result has been deallocated as that will likely cause a segmentation fault or other severe failure.

This method exists to make it easier to work with C extensions without going through NumPy. If you want to pass the search results to NumPy then use as_numpy_array() instead.

Returns:a ctypes array of type Hit*len(self)
as_numpy_array()

Return a NumPy array view of the underlying indices and scores

The view uses a structured types with fields ‘index’ (i4) and ‘score’ (f8), mapped directly onto chemfp’s own data structure. For example, to get the score of the 4th entry use:

result.as_numpy_array()["score"][3]
    -or-
result.as_numpy_array()[3][1]

Modifications to this view will change chemfp’s data values and vice versa. USE WITH CARE!

Warning

Do not attempt to access the NumPy array contents after the search result has been deallocated as that will likely cause a segmentation fault or other severe failure.

As a short-hand to get just the indices or just the scores, use get_indices_as_numpy_array() or get_scores_as_numpy_array().

Returns:a NumPy array with a structured data type
count(min_score=None, max_score=None, interval='[]')

Count the number of hits with a score between min_score and max_score

Using the default parameters this returns the number of hits in the result.

The default min_score of None is equivalent to -infinity. The default max_score of None is equivalent to +infinity.

The interval parameter describes the interval end conditions. The default of “[]” uses a closed interval, where min_score <= score <= max_score. The interval “()” uses the open interval where min_score < score < max_score. The half-open/half-closed intervals “(]” and “[)” are also supported.

Parameters:
  • min_score (a float, or None for -infinity) – the minimum score in the range.
  • max_score (a float, or None for +infinity) – the maximum score in the range.
  • interval (one of "[]", "()", "(]", "[)") – specify if the end points are open or closed.
Returns:

an integer count

cumulative_score(min_score=None, max_score=None, interval='[]')

The sum of the scores which are between min_score and max_score

Using the default parameters this returns the sum of all of the scores in the result. With a specified range this returns the sum of all of the scores in that range. The cumulative score is also known as the raw score.

The default min_score of None is equivalent to -infinity. The default max_score of None is equivalent to +infinity.

The interval parameter describes the interval end conditions. The default of “[]” uses a closed interval, where min_score <= score <= max_score. The interval “()” uses the open interval where min_score < score < max_score. The half-open/half-closed intervals “(]” and “[)” are also supported.

Parameters:
  • min_score (a float, or None for -infinity) – the minimum score in the range.
  • max_score (a float, or None for +infinity) – the maximum score in the range.
  • interval (one of "[]", "()", "(]", "[)") – specify if the end points are open or closed.
Returns:

a floating point value

format_ids_and_scores_as_bytes(ids=None, precision=4)

Format the ids and scores as the byte string needed for simsearch output

If there are no hits then the result is the empty string b””, otherwise it returns a byte string containing the tab-seperated ids and scores, in the order ids[0], scores[0], ids[1], scores[1], …

If the ids is not specified then the ids come from self.get_ids(). If no ids are available, a ValueError is raised. The ids must be a list of Unicode strings.

The precision sets the number of decimal digits to use in the score output. It must be an integer value between 1 and 10, inclusive.

This function is 3-4x faster than the Python equivalent, which is roughly:

ids = ids if (ids is not None) else self.get_ids()
formatter = ("%s\t%." + str(precision) + "f").encode("ascii")
return b"\t".join(formatter % pair for pair in zip(ids, self.get_scores()))
Parameters:
  • ids (a list of Unicode strings, or None to use the default) – the identifiers to use for each hit.
  • precision (an integer from 1 to 10, inclusive) – the precision to use for each score
Returns:

a byte string

get_ids()

The list of target identifiers (if available), in the current ordering

Returns:a list of strings
get_ids_and_scores()

The list of (target identifier, target score) pairs, in the current ordering

Raises a TypeError if the target IDs are not available.

Returns:a Python list of 2-element tuples
get_indices()

The list of target indices, in the current ordering.

This returns a copy of the scores. See get_indices_as_numpy_array() to get a NumPy array view of the indices.

Returns:an array.array() of type ‘i’
get_indices_and_scores()

The list of (target index, target score) pairs, in the current ordering

Returns:a Python list of 2-element tuples
get_indices_as_numpy_array()

Return a NumPy array view of the underlying indices.

This is a short-cut for self.as_numpy_array()[“index”]. See that method documentation for details and warning.

Returns:a NumPy array of type ‘i4’
get_scores()

The list of target scores, in the current ordering

This returns a copy of the scores. See get_scores_as_numpy_array() to get a NumPy array view of the scores.

Returns:an array.array() of type ‘d’
get_scores_as_numpy_array()

Return a NumPy array view of the underlying scores.

This is a short-cut for self.as_numpy_array()[“score”]. See that method documentation for details and warning.

Returns:a NumPy array of type ‘f8’
iter_ids()

Iterate over target identifiers (if available), in the current ordering

max()

Return the value of the largest score

Returns 0.0 if there are no results.

Returns:a float
min()

Return the value of the smallest score

Returns 0.0 if there are no results.

Returns:a float
query_id

Return the corresponding query id, if available, else None

reorder(ordering='decreasing-score-plus')

Reorder the hits based on the requested ordering.

The available orderings are:

  • increasing-score - sort by increasing score
  • decreasing-score - sort by decreasing score
  • increasing-score-plus - sort by increasing score, break ties by increasing index
  • decreasing-score-plus - sort by decreasing score, break ties by increasing index
  • increasing-index - sort by increasing target index
  • decreasing-index - sort by decreasing target index
  • move-closest-first - move the hit with the highest score to the first position
  • reverse - reverse the current ordering
Parameters:ordering (string) – the name of the ordering to use
save(destination, format=None, compressed=True)

Save the SearchResult to the given destination

Currently only the “npz” format is supported, which is a NumPy format containing multiple arrays, each stored as a file entry in a zipfile. The SearchResult is stored in the same structure as a SciPy compressed sparse row (‘csr’) matrix, which means they can be read with scipy.sparse.load_npz().

Chemfp also stores the query and target identifiers in the npz file, the chemfp search parameters like the number of bits in the fingerprint or the values for alpha and beta, and a value indicating the array contains a single SearchResult.

Use chemfp.search.load_npz() to read the similarity search result back into a SearchResult instance.

Parameters:
  • destination (a filename, binary file object, or None for stdout) – where to write the results
  • format (None or 'npz') – the output format name (default: always ‘npz’)
  • compressed – if True (the default), use zipfile compression
target_ids

Return all the original target ids, not just the hit ids for this query

target_start

Return the start offset of the hit into the original arena

to_pandas(*, columns=['target_id', 'score'])

Return a pandas DataFrame with the target ids and scores

The first column contains the ids, the second column contains the ids. The default columns headers are “target_id” and “score”. Use columns to specify different headers.

Parameters:columns (a list of two strings) – column names for the returned DataFrame
Returns:a pandas DataFrame
class chemfp.search.SearchResults(num_rows, num_cols, query_arena=None, query_ids=None, target_arena=None, target_arena_ids=None, target_start=0, num_bits=2147483647, alpha=1.0, beta=1.0, matrix_type=<MatrixType.NxM: 0>)

Bases: chemfp.search.SearchResults

Search results for a list of query fingerprints against a target arena

This acts like a list of SearchResult elements, with the ability to iterate over each search results, look them up by index, and get the number of scores.

In addition, there are helper methods to iterate over each hit and to get the hit indicies, scores, and identifiers directly as Python lists, sort the list contents, and more.

query_ids

A list of query ids, one for each result. This comes from the query arena’s ids.

count_all(min_score=None, max_score=None, interval='[]')

Count the number of hits with a score between min_score and max_score

Using the default parameters this returns the number of hits in the result.

The default min_score of None is equivalent to -infinity. The default max_score of None is equivalent to +infinity.

The interval parameter describes the interval end conditions. The default of “[]” uses a closed interval, where min_score <= score <= max_score. The interval “()” uses the open interval where min_score < score < max_score. The half-open/half-closed intervals “(]” and “[)” are also supported.

Parameters:
  • min_score (a float, or None for -infinity) – the minimum score in the range.
  • max_score (a float, or None for +infinity) – the maximum score in the range.
  • interval (one of "[]", "()", "(]", "[)") – specify if the end points are open or closed.
Returns:

an integer count

cumulative_score_all(min_score=None, max_score=None, interval='[]')

The sum of all scores in all rows which are between min_score and max_score

Using the default parameters this returns the sum of all of the scores in all of the results. With a specified range this returns the sum of all of the scores in that range. The cumulative score is also known as the raw score.

The default min_score of None is equivalent to -infinity. The default max_score of None is equivalent to +infinity.

The interval parameter describes the interval end conditions. The default of “[]” uses a closed interval, where min_score <= score <= max_score. The interval “()” uses the open interval where min_score < score < max_score. The half-open/half-closed intervals “(]” and “[)” are also supported.

Parameters:
  • min_score (a float, or None for -infinity) – the minimum score in the range.
  • max_score (a float, or None for +infinity) – the maximum score in the range.
  • interval (one of "[]", "()", "(]", "[)") – specify if the end points are open or closed.
Returns:

a floating point count

iter_ids()

For each hit, yield the list of target identifiers

iter_ids_and_scores()

For each hit, yield the list of (target id, score) tuples

iter_indices()

For each hit, yield the list of target indices

iter_indices_and_scores()

For each hit, yield the list of (target index, score) tuples

iter_scores()

For each hit, yield the list of target scores

matrix_type

Get the MatrixType enum value, or an integer if out of range

matrix_type_name

Get a string describing the matrix type, or ‘unknown’ if for unknown types

reorder_all(ordering='decreasing-score-plus')

Reorder the hits for all of the rows based on the requested order.

The available orderings are:

  • increasing-score - sort by increasing score
  • decreasing-score - sort by decreasing score
  • increasing-score-plus - sort by increasing score, break ties by increasing index
  • decreasing-score-plus - sort by decreasing score, break ties by increasing index
  • increasing-index - sort by increasing target index
  • decreasing-index - sort by decreasing target index
  • move-closest-first - move the hit with the highest score to the first position
  • reverse - reverse the current ordering
Parameters:ordering (string) – the name of the ordering to use
save(destination, format=None, compressed=True)

Save the SearchResults to the given destination

Currently only the “npz” format is supported, which is a NumPy format containing multiple arrays, each stored as a file entry in a zipfile. The SearchResults results are stored in the same structure as a SciPy compressed sparse row (‘csr’) matrix, which means they can be read with scipy.sparse.load_npz().

Chemfp also stores the query and target identifiers in the npz file, the chemfp search parameters like the number of bits in the fingerprint or the values for alpha and beta, and a value indicating the array contains a SearchResults.

Use chemfp.search.load_npz() to read the similarity search results back into a SearchResults instance.

Parameters:
  • destination (a filename, binary file object, or None for stdout) – where to write the results
  • format (None or 'npz') – the output format name (default: always ‘npz’)
  • compressed – if True (the default), use zipfile compression
shape

the tuple (number of rows, number of columns)

The number of columns is the size of the target arena.

target_ids

The target ids

Be aware that these indices always start from 0 and go up to the number of columns, while the indices from get_indices() and related functions are relative to the initial position in the target arena. If the targets were a subarena slice which did not start from 0 then the indices will differ by target_start.

Use target_arena_ids to get the ids indexed across their full range.

to_csr(dtype=None)

Return the results as a SciPy compressed sparse row matrix.

The returned matrix has the same shape as the SearchResult instance and can be passed into, for example, a scikit-learn clustering algorithm.

By default the scores are stored with the dtype of “float64”. You may also use “float32” though mind the double rounding from int/int -> double -> float.

This method requires that SciPy (and NumPy) be installed.

Parameters:dtype (string or NumPy type, or None for float64) – a NumPy numeric data type (either “float64” or “float32”)
to_pandas(*, columns=['query_id', 'target_id', 'score'], empty=('*', None))

Return a pandas DataFrame with query_id, target_id and score columns

Each query has zero or more hits. Each hit becomes a row in the output table, with the query id in the first column, the hit target id in the second, and the hit score in the third.

If a query has no hits then by default a row is added with the query id, ‘*’ as the target id, and None as the score (which pandas will treat as a NA value).

Use empty to specify different behavior for queries with no hits. If empty is None then no row is added to the table. If empty is a 2-element tuple the first element is used as the target id and the second is used as the score.

Use the DataFrame’s groupby() method to group results by query id, for example:

>>> import chemfp
>>> df = chemfp.simsearch(queries="queries.fps", targets="targets.fps",
...        k=10, threshold=0.4, progress=False).to_pandas()
>>> df.groupby("query_id").describe()
Parameters:
  • columns (a list of three strings) – column names for the returned DataFrame
  • empty (a list of two strings, or None) – the target id and score used for queries with no hits, or None to not include a row for that case
Returns:

a pandas DataFrame

chemfp.search.get_search_results(obj)

Convert the given object to a SearchResults, or raise a TypeError

This is a functools.singledispatch registry hook used so the high-level simsearch() result can be passed in as a SearchResults.

chemfp.search.from_csr(matrix, query_ids=None, target_ids=None, num_bits=2147483647, alpha=1.0, beta=1.0, matrix_type=<MatrixType.NxM: 0>)

Convert a SciPy compressed sparse row matrix to a SearchResults

If not specified then query_ids will be list(range(num_rows)) and target_ids will be list(range(num_cols)). The `target_ids must be at least as long as the maximum index in the sparse matrix.

The num_bits, alpha and beta must match the parameters used to generate the Tanimoto or Tversky scores otherwise some of the chemfp internal optimizations may give incorrect sort results. Keep the default values if you do not know the parameters, or used another method to generate the scores.

matrix_type describes the contents of the matrix. Can be “NxM” or “no-diagonal”. The latter is only valid if the target_ids are the same object (meaning the queries and targets are the same), and if the diagonal term is not included in the matrix.

The scores must not contain infinite numbers or NaNs.

Parameters:
  • matrix – a SciPy compressed sparse row (“csr”) matrix
  • query_ids – a list of identifiers for each row
  • target_ids – a list of identifiers, used to match index to target id
  • num_bits (an integer between 1 and 2**31-1) – the number of fingerprint bits, or a larger value
  • alpha (a float between 0.0 and 10.0) – the Tversky alpha value
  • beta (a float between 0.0 and 10.0) – the Tversky beta value
  • matrix_type (either 'NxM' or 'no-diagonal') – a description of the matrix contents
Returns:

a SearchResults.

chemfp.search.count_tanimoto_hits_fp(query_fp, target_arena, threshold=0.7)

Count the number of hits in target_arena at least threshold similar to the query_fp

Example:

query_id, query_fp = chemfp.load_fingerprints("queries.fps")[0]
targets = chemfp.load_fingerprints("targets.fps")
print(chemfp.search.count_tanimoto_hits_fp(query_fp, targets, threshold=0.1))
Parameters:
  • query_fp (a byte string) – the query fingerprint
  • target_arena (a FingerprintArena) – the target arena
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
Returns:

an integer count

chemfp.search.count_tanimoto_hits_arena(query_arena, target_arena, threshold=0.7)

For each fingerprint in query_arena, count the number of hits in target_arena at least threshold similar to it

Example:

queries = chemfp.load_fingerprints("queries.fps")
targets = chemfp.load_fingerprints("targets.fps")
counts = chemfp.search.count_tanimoto_hits_arena(queries, targets, threshold=0.1)
print(counts[:10])

The result is implementation specific. You’ll always be able to get its length and do an index lookup to get an integer count. Currently it’s a ctypes array of longs, but it could be an array.array or Python list in the future.

Parameters:
Returns:

an array of counts

chemfp.search.count_tanimoto_hits_symmetric(arena, threshold=0.7, *, batch_size=1000, batch_callback=None)

For each fingerprint in the arena, count the number of other fingerprints at least threshold similar to it

A fingerprint never matches itself.

The computation can take a long time. Python won’t check check for a ^C until the function finishes. This can be irritating. Instead, process only batch_size rows at a time before checking for a ^C.

Example:

arena = chemfp.load_fingerprints("targets.fps")
counts = chemfp.search.count_tanimoto_hits_symmetric(arena, threshold=0.2)
print(counts[:10])

The result object is implementation specific. You’ll always be able to get its length and do an index lookup to get an integer count. Currently it’s a ctype array of longs, but it could be an array.array or Python list in the future.

Parameters:
  • arena (a chemfp.arena.FingerprintArena) – the set of fingerprints
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • batch_size (integer) – the number of rows to process before checking for a ^C
Returns:

an array of counts

chemfp.search.partial_count_tanimoto_hits_symmetric(counts, arena, threshold=0.7, query_start=0, query_end=None, target_start=0, target_end=None)

Compute a portion of the symmetric Tanimoto counts

For most cases, use chemfp.search.count_tanimoto_hits_symmetric() instead of this function!

This function is only useful for thread-pool implementations. In that case, set the number of OpenMP threads to 1.

counts is a contiguous array of integers. It should be initialized to zeros, and reused for successive calls.

The function adds counts for counts[query_start:query_end] based on computing the upper-triangle portion contained in the rectangle query_start:query_end and target_start:target_end* and using symmetry to fill in the lower half.

You know, this is pretty complicated. Here’s the bare minimum example of how to use it correctly to process 10 rows at a time using up to 4 threads:

import chemfp
import chemfp.search
from chemfp import futures
import array

chemfp.set_num_threads(1)  # Globally disable OpenMP

arena = chemfp.load_fingerprints("targets.fps")  # Load the fingerprints
n = len(arena)
counts = array.array("i", [0]*n)

with futures.ThreadPoolExecutor(max_workers=4) as executor:
    for row in range(0, n, 10):
        executor.submit(chemfp.search.partial_count_tanimoto_hits_symmetric,
                        counts, arena, threshold=0.2,
                        query_start=row, query_end=min(row+10, n))

print(counts)
Parameters:
  • counts (a contiguous block of integer) – the accumulated Tanimoto counts
  • arena (a chemfp.arena.FingerprintArena) – the fingerprints.
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • query_start (an integer) – the query start row
  • query_end (an integer, or None to mean the last query row) – the query end row
  • target_start (an integer) – the target start row
  • target_end (an integer, or None to mean the last target row) – the target end row
Returns:

None

chemfp.search.threshold_tanimoto_search_fp(query_fp, target_arena, threshold=0.7)

Search for fingerprint hits in target_arena which are at least threshold similar to query_fp

The hits in the returned chemfp.search.SearchResult are in arbitrary order.

Example:

query_id, query_fp = chemfp.load_fingerprints("queries.fps")[0]
targets = chemfp.load_fingerprints("targets.fps")
print(list(chemfp.search.threshold_tanimoto_search_fp(query_fp, targets, threshold=0.15)))
Parameters:
  • query_fp (a byte string) – the query fingerprint
  • target_arena (a chemfp.arena.FingerprintArena) – the target arena
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
Returns:

a chemfp.search.SearchResult

chemfp.search.threshold_tanimoto_search_arena(query_arena, target_arena, threshold=0.7, batch_size=None, batch_callback=None)

Search for the hits in the target_arena at least threshold similar to the fingerprints in query_arena

The hits in the returned chemfp.search.SearchResults are in arbitrary order.

Example:

queries = chemfp.load_fingerprints("queries.fps")
targets = chemfp.load_fingerprints("targets.fps")
results = chemfp.search.threshold_tanimoto_search_arena(queries, targets, threshold=0.5)
for query_id, query_hits in zip(queries.ids, results):
    if len(query_hits) > 0:
        print(query_id, "->", ", ".join(query_hits.get_ids()))
Parameters:
Returns:

a chemfp.search.SearchResults

chemfp.search.threshold_tanimoto_search_symmetric(arena, threshold=0.7, include_lower_triangle=True, batch_size=1000, batch_callback=None)

Search for the hits in the arena at least threshold similar to the fingerprints in the arena

When include_lower_triangle is True, compute the upper-triangle similarities, then copy the results to get the full set of results. When include_lower_triangle is False, only compute the upper triangle.

The hits in the returned chemfp.search.SearchResults are in arbitrary order.

The computation can take a long time. Python won’t check check for a ^C until the function finishes. This can be irritating. Instead, process only batch_size rows at a time before checking for a ^C.

Example:

arena = chemfp.load_fingerprints("queries.fps")
full_result = chemfp.search.threshold_tanimoto_search_symmetric(arena, threshold=0.2)
upper_triangle = chemfp.search.threshold_tanimoto_search_symmetric(
          arena, threshold=0.2, include_lower_triangle=False)
assert sum(map(len, full_result)) == sum(map(len, upper_triangle))*2
Parameters:
  • arena (a chemfp.arena.FingerprintArena) – the set of fingerprints
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • include_lower_triangle (boolean) – if False, compute only the upper triangle, otherwise use symmetry to compute the full matrix
  • batch_size (integer) – the number of rows to process before checking for a ^C
Returns:

a chemfp.search.SearchResults

chemfp.search.partial_threshold_tanimoto_search_symmetric(results, arena, threshold=0.7, query_start=0, query_end=None, target_start=0, target_end=None, results_offset=0)

Compute a portion of the symmetric Tanimoto search results

For most cases, use chemfp.search.threshold_tanimoto_search_symmetric() instead of this function!

This function is only useful for thread-pool implementations. In that case, set the number of OpenMP threads to 1.

results is a chemfp.search.SearchResults instance which is at least as large as the arena. It should be reused for successive updates.

The function adds hits to results[query_start:query_end], based on computing the upper-triangle portion contained in the rectangle query_start:query_end and target_start:target_end.

It does not fill in the lower triangle. To get the full matrix, call fill_lower_triangle.

You know, this is pretty complicated. Here’s the bare minimum example of how to use it correctly to process 10 rows at a time using up to 4 threads:

import chemfp
from chemfp.search import (SearchResults,
    MatrixType, partial_threshold_tanimoto_search_symmetric)
from concurrent.futures import ThreadPoolExecutor

chemfp.set_num_threads(1)

arena = chemfp.load_fingerprints("targets.fps")
n = len(arena)
results = SearchResults(n, n, query_ids=arena.ids, target_ids=arena.ids,
                        matrix_type=MatrixType.NO_DIAGONAL)

with ThreadPoolExecutor(max_workers=4) as executor:
    for row in range(0, n, 10):
        executor.submit(
          partial_threshold_tanimoto_search_symmetric,
          results, arena, threshold=0.2,
          query_start=row, query_end=min(row+10, n))

chemfp.search.fill_lower_triangle(results)

The hits in the chemfp.search.SearchResults are in arbitrary order.

Parameters:
  • results (a chemfp.search.SearchResults instance) – the intermediate search results
  • arena (a chemfp.arena.FingerprintArena) – the fingerprints.
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • query_start (an integer) – the query start row
  • query_end (an integer, or None to mean the last query row) – the query end row
  • target_start (an integer) – the target start row
  • target_end (an integer, or None to mean the last target row) – the target end row
  • results_offset – use results[results_offset] as the base for the results
  • results_offset – an integer
Returns:

None

chemfp.search.fill_lower_triangle(results)

Duplicate each entry of results to its transpose

This is used after the symmetric threshold search to turn the upper-triangle results into a full matrix.

Parameters:results (a chemfp.search.SearchResults) – search results
chemfp.search.knearest_tanimoto_search_fp(query_fp, target_arena, k=3, threshold=0.0)

Search for k-nearest hits in target_arena which are at least threshold similar to query_fp

The hits in the chemfp.search.SearchResult are ordered by decreasing similarity score.

Example:

query_id, query_fp = chemfp.load_fingerprints("queries.fps")[0]
targets = chemfp.load_fingerprints("targets.fps")
print(list(chemfp.search.knearest_tanimoto_search_fp(query_fp, targets, k=3, threshold=0.0)))
Parameters:
  • query_fp (a byte string) – the query fingerprint
  • target_arena (a chemfp.arena.FingerprintArena) – the target arena
  • k (positive integer) – the number of nearest neighbors to find.
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
Returns:

a chemfp.search.SearchResult

chemfp.search.knearest_tanimoto_search_arena(query_arena, target_arena, k=3, threshold=0.0, query_thresholds=None, batch_size=None, batch_callback=None)

Search for the k nearest hits in the target_arena at least threshold similar to the fingerprints in query_arena

The hits in the chemfp.search.SearchResults are ordered by decreasing similarity score.

Example:

queries = chemfp.load_fingerprints("queries.fps")
targets = chemfp.load_fingerprints("targets.fps")
results = chemfp.search.knearest_tanimoto_search_arena(queries, targets, k=3, threshold=0.5)
for query_id, query_hits in zip(queries.ids, results):
    if len(query_hits) >= 2:
        print(query_id, "->", ", ".join(query_hits.get_ids()))

Use query_thresholds to specify per-query thresholds instead of using the global threshold. The global threshold must still be in range 0.0 to 1.0.

Parameters:
  • query_arena (a chemfp.arena.FingerprintArena) – The query fingerprints.
  • target_arena (a chemfp.arena.FingerprintArena) – The target fingerprints.
  • k (positive integer) – the number of nearest neighbors to find.
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • query_thresholds (None or a list of Python floats, or an array of C doubles) – optionally specify per-query thresholds
Returns:

a chemfp.search.SearchResults

chemfp.search.knearest_tanimoto_search_symmetric(arena, k=3, threshold=0.0, query_thresholds=None, batch_size=1000, batch_callback=None)

Search for the k-nearest hits in the arena at least threshold similar to the fingerprints in the arena

The hits in the SearchResults are ordered by decreasing similarity score.

The computation can take a long time. Python won’t check check for a ^C until the function finishes. This can be irritating. Instead, process only batch_size rows at a time before checking for a ^C.

Example:

arena = chemfp.load_fingerprints("queries.fps")
results = chemfp.search.knearest_tanimoto_search_symmetric(arena, k=3, threshold=0.8)
for (query_id, hits) in zip(arena.ids, results):
    print(query_id, "->", ", ".join(("%s %.2f" % hit) for hit in  hits.get_ids_and_scores()))
Parameters:
  • arena (a chemfp.arena.FingerprintArena) – the set of fingerprints
  • k (positive integer) – the number of nearest neighbors to find.
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • query_thresholds (None or a list of Python floats, or an array of C doubles) – optionally specify per-query thresholds
  • batch_size (integer) – the number of rows to process before checking for a ^C
Returns:

a chemfp.search.SearchResults

chemfp.search.count_tversky_hits_fp(query_fp, target_arena, threshold=0.7, alpha=1.0, beta=1.0)

Count the number of hits in target_arena least threshold similar to the query_fp (Tversky)

Example:

query_id, query_fp = chemfp.load_fingerprints("queries.fps")[0]
targets = chemfp.load_fingerprints("targets.fps")
print(chemfp.search.count_tversky_hits_fp(query_fp, targets, threshold=0.1))
Parameters:
  • query_fp (a byte string) – the query fingerprint
  • target_arena – the target arena
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • alpha (a value between 0.0 and 100.0, inclusive) – the Tversky alpha value
  • beta (a value between 0.0 and 100.0, inclusive) – the Tversky beta value
Returns:

an integer count

chemfp.search.count_tversky_hits_arena(query_arena, target_arena, threshold=0.7, alpha=1.0, beta=1.0)

For each fingerprint in query_arena, count the number of hits in target_arena at least threshold similar to it

Example:

queries = chemfp.load_fingerprints("queries.fps")
targets = chemfp.load_fingerprints("targets.fps")
counts = chemfp.search.count_tversky_hits_arena(queries, targets, threshold=0.1,
              alpha=0.5, beta=0.5)
print(counts[:10])

The result is implementation specific. You’ll always be able to get its length and do an index lookup to get an integer count. Currently it’s a ctypes array of longs, but it could be an array.array or Python list in the future.

Parameters:
  • query_arena (a chemfp.arena.FingerprintArena) – The query fingerprints.
  • target_arena (a chemfp.arena.FingerprintArena) – The target fingerprints.
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • alpha (a value between 0.0 and 100.0, inclusive) – the Tversky alpha value
  • beta (a value between 0.0 and 100.0, inclusive) – the Tversky beta value
Returns:

an array of counts

chemfp.search.count_tversky_hits_symmetric(arena, threshold=0.7, alpha=1.0, beta=1.0, batch_size=1000, batch_callback=None)

For each fingerprint in the arena, count the number of other fingerprints at least threshold similar to it

A fingerprint never matches itself.

The computation can take a long time. Python won’t check check for a ^C until the function finishes. This can be irritating. Instead, process only batch_size rows at a time before checking for a ^C.

Example:

arena = chemfp.load_fingerprints("targets.fps")
counts = chemfp.search.count_tversky_hits_symmetric(
      arena, threshold=0.2, alpha=0.5, beta=0.5)
print(counts[:10])

The result object is implementation specific. You’ll always be able to get its length and do an index lookup to get an integer count. Currently it’s a ctype array of longs, but it could be an array.array or Python list in the future.

Parameters:
  • arena (a chemfp.arena.FingerprintArena) – the set of fingerprints
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • alpha (a value between 0.0 and 100.0, inclusive) – the Tversky alpha value
  • beta (a value between 0.0 and 100.0, inclusive) – the Tversky beta value
  • batch_size (integer) – the number of rows to process before checking for a ^C
Returns:

an array of counts

chemfp.search.partial_count_tversky_hits_symmetric(counts, arena, threshold=0.7, alpha=1.0, beta=1.0, query_start=0, query_end=None, target_start=0, target_end=None)

Compute a portion of the symmetric Tversky counts

For most cases, use chemfp.search.count_tversky_hits_symmetric() instead of this function!

This function is only useful for thread-pool implementations. In that case, set the number of OpenMP threads to 1.

counts is a contiguous array of integers. It should be initialized to zeros, and reused for successive calls.

The function adds counts for counts[query_start:query_end] based on computing the upper-triangle portion contained in the rectangle query_start:query_end and target_start:target_end* and using symmetry to fill in the lower half.

You know, this is pretty complicated. Here’s the bare minimum example of how to use it correctly to process 10 rows at a time using up to 4 threads:

import chemfp
import chemfp.search
from chemfp import futures
import array

chemfp.set_num_threads(1)  # Globally disable OpenMP

arena = chemfp.load_fingerprints("targets.fps")  # Load the fingerprints
n = len(arena)
counts = array.array("i", [0]*n)

with futures.ThreadPoolExecutor(max_workers=4) as executor:
    for row in range(0, n, 10):
        executor.submit(chemfp.search.partial_count_tversky_hits_symmetric,
                        counts, arena, threshold=0.2, alpha=0.5, beta=0.5,
                        query_start=row, query_end=min(row+10, n))

print(counts)
Parameters:
  • counts (a contiguous block of integer) – the accumulated Tversky counts
  • arena (a chemfp.arena.FingerprintArena) – the fingerprints.
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • alpha (a value between 0.0 and 100.0, inclusive) – the Tversky alpha value
  • beta (a value between 0.0 and 100.0, inclusive) – the Tversky beta value
  • query_start (an integer) – the query start row
  • query_end (an integer, or None to mean the last query row) – the query end row
  • target_start (an integer) – the target start row
  • target_end (an integer, or None to mean the last target row) – the target end row
Returns:

None

chemfp.search.threshold_tversky_search_fp(query_fp, target_arena, threshold=0.7, alpha=1.0, beta=1.0)

Search for fingerprint hits in target_arena which are at least threshold similar to query_fp

The hits in the returned chemfp.search.SearchResult are in arbitrary order.

Example:

query_id, query_fp = chemfp.load_fingerprints("queries.fps")[0]
targets = chemfp.load_fingerprints("targets.fps")
print(list(chemfp.search.threshold_tversky_search_fp(
           query_fp, targets, threshold=0.15, alpha=0.5, beta=0.5)))
Parameters:
  • query_fp (a byte string) – the query fingerprint
  • target_arena (a chemfp.arena.FingerprintArena) – the target arena
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • alpha (a value between 0.0 and 100.0, inclusive) – the Tversky alpha value
  • beta (a value between 0.0 and 100.0, inclusive) – the Tversky beta value
Returns:

a chemfp.search.SearchResult

chemfp.search.threshold_tversky_search_arena(query_arena, target_arena, threshold=0.7, alpha=1.0, beta=1.0, batch_size=None, batch_callback=None)

Search for the hits in the target_arena at least threshold similar to the fingerprints in query_arena

The hits in the returned chemfp.search.SearchResults are in arbitrary order.

Example:

queries = chemfp.load_fingerprints("queries.fps")
targets = chemfp.load_fingerprints("targets.fps")
results = chemfp.search.threshold_tversky_search_arena(
              queries, targets, threshold=0.5, alpha=0.5, beta=0.5)
for query_id, query_hits in zip(queries.ids, results):
    if len(query_hits) > 0:
        print(query_id, "->", ", ".join(query_hits.get_ids()))
Parameters:
  • query_arena (a chemfp.arena.FingerprintArena) – The query fingerprints.
  • target_arena (a chemfp.arena.FingerprintArena) – The target fingerprints.
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • alpha (a value between 0.0 and 100.0, inclusive) – the Tversky alpha value
  • beta (a value between 0.0 and 100.0, inclusive) – the Tversky beta value
Returns:

a chemfp.search.SearchResults

chemfp.search.threshold_tversky_search_symmetric(arena, threshold=0.7, alpha=1.0, beta=1.0, include_lower_triangle=True, batch_size=1000, batch_callback=None)

Search for the hits in the arena at least threshold similar to the fingerprints in the arena

When include_lower_triangle is True, compute the upper-triangle similarities, then copy the results to get the full set of results. When include_lower_triangle is False, only compute the upper triangle.

The hits in the returned chemfp.search.SearchResults are in arbitrary order.

The computation can take a long time. Python won’t check check for a ^C until the function finishes. This can be irritating. Instead, process only batch_size rows at a time before checking for a ^C

Example:

arena = chemfp.load_fingerprints("queries.fps")
full_result = chemfp.search.threshold_tversky_search_symmetric(
      arena, threshold=0.2, alpha=0.5, beta=0.5)
upper_triangle = chemfp.search.threshold_tversky_search_symmetric(
          arena, threshold=0.2, alpha=0.5, beta=0.5, include_lower_triangle=False)
assert sum(map(len, full_result)) == sum(map(len, upper_triangle))*2
Parameters:
  • arena (a chemfp.arena.FingerprintArena) – the set of fingerprints
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • alpha (a value between 0.0 and 100.0, inclusive) – the Tversky alpha value
  • beta (a value between 0.0 and 100.0, inclusive) – the Tversky beta value
  • include_lower_triangle (boolean) – if False and alpha == beta, compute only the upper triangle, otherwise use symmetry to compute the full matrix
  • batch_size (integer) – the number of rows to process before checking for a ^C
Returns:

a chemfp.search.SearchResults

chemfp.search.partial_threshold_tversky_search_symmetric(results, arena, threshold=0.7, alpha=1.0, beta=1.0, query_start=0, query_end=None, target_start=0, target_end=None, results_offset=0)

Compute a portion of the symmetric Tversky search results

For most cases, use chemfp.search.threshold_tversky_search_symmetric() instead of this function!

This function is only useful for thread-pool implementations. In that case, set the number of OpenMP threads to 1.

results is a chemfp.search.SearchResults instance which is at least as large as the arena. It should be reused for successive updates.

The function adds hits to results[query_start:query_end], based on computing the upper-triangle portion contained in the rectangle query_start:query_end and target_start:target_end.

It does not fill in the lower triangle. To get the full matrix, call fill_lower_triangle.

You know, this is pretty complicated. Here’s the bare minimum example of how to use it correctly to process 10 rows at a time using up to 4 threads:

import chemfp
from chemfp.search import (SearchResults, MatrixType,
    partial_threshold_tversky_search_symmetric)
from concurrent.futures import ThreadPoolExecutor

chemfp.set_num_threads(1)

arena = chemfp.load_fingerprints("targets.fps")
n = len(arena)
results = SearchResults(
        n, n, query_ids=arena.ids, target_ids=arena.ids,
        matrix_type = MatrixType.STRICTLY_UPPER_TRIANGULAR)

with ThreadPoolExecutor(max_workers=4) as executor:
    for row in range(0, n, 10):
        executor.submit(
            partial_threshold_tversky_search_symmetric,
            results, arena, threshold=0.2, alpha=0.5, beta=0.5,
            query_start=row, query_end=min(row+10, n))

chemfp.search.fill_lower_triangle(results)

The hits in the chemfp.search.SearchResults are in arbitrary order.

Parameters:
  • counts (an SearchResults instance) – the intermediate search results
  • arena (a chemfp.arena.FingerprintArena) – the fingerprints.
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • alpha (a value between 0.0 and 100.0, inclusive) – the Tversky alpha value
  • beta (a value between 0.0 and 100.0, inclusive) – the Tversky beta value
  • query_start (an integer) – the query start row
  • query_end (an integer, or None to mean the last query row) – the query end row
  • target_start (an integer) – the target start row
  • target_end (an integer, or None to mean the last target row) – the target end row
  • results_offset – use results[results_offset] as the base for the results
  • results_offset – an integer
Returns:

None

chemfp.search.knearest_tversky_search_fp(query_fp, target_arena, k=3, threshold=0.0, alpha=1.0, beta=1.0)

Search for k-nearest hits in target_arena which are at least threshold similar to query_fp

The hits in the chemfp.search.SearchResult are ordered by decreasing similarity score.

Example:

query_id, query_fp = chemfp.load_fingerprints("queries.fps")[0]
targets = chemfp.load_fingerprints("targets.fps")
print(list(chemfp.search.knearest_tversky_search_fp(
        query_fp, targets, k=3, threshold=0.0, alpha=0.5, beta=0.5)))
Parameters:
  • query_fp (a byte string) – the query fingerprint
  • target_arena – the target arena
  • k (positive integer) – the number of nearest neighbors to find.
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • alpha (a value between 0.0 and 100.0, inclusive) – the Tversky alpha value
  • beta (a value between 0.0 and 100.0, inclusive) – the Tversky beta value
Returns:

a chemfp.search.SearchResult

chemfp.search.knearest_tversky_search_arena(query_arena, target_arena, k=3, threshold=0.0, alpha=1.0, beta=1.0, query_thresholds=None, batch_size=None, batch_callback=None)

Search for the k nearest hits in the target_arena at least threshold similar to the fingerprints in query_arena

The hits in the chemfp.search.SearchResults are ordered by decreasing similarity score.

Example:

queries = chemfp.load_fingerprints("queries.fps")
targets = chemfp.load_fingerprints("targets.fps")
results = chemfp.search.knearest_tversky_search_arena(
      queries, targets, k=3, threshold=0.5, alpha=0.5, beta=0.5)
for query_id, query_hits in zip(queries.ids, results):
    if len(query_hits) >= 2:
        print(query_id, "->", ", ".join(query_hits.get_ids()))
Parameters:
  • query_arena (a chemfp.arena.FingerprintArena) – The query fingerprints.
  • target_arena (a chemfp.arena.FingerprintArena) – The target fingerprints.
  • k (positive integer) – the number of nearest neighbors to find.
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • alpha (a value between 0.0 and 100.0, inclusive) – the Tversky alpha value
  • beta (a value between 0.0 and 100.0, inclusive) – the Tversky beta value
Returns:

a chemfp.search.SearchResults

chemfp.search.knearest_tversky_search_symmetric(arena, k=3, threshold=0.0, alpha=1.0, beta=1.0, query_thresholds=None, batch_size=1000, batch_callback=None)

Search for the k-nearest hits in the arena at least threshold similar to the fingerprints in the arena

The hits in the SearchResults are ordered by decreasing similarity score.

The computation can take a long time. Python won’t check check for a ^C until the function finishes. This can be irritating. Instead, process only batch_size rows at a time before checking for a ^C.

Example:

arena = chemfp.load_fingerprints("queries.fps")
results = chemfp.search.knearest_tversky_search_symmetric(
         arena, k=3, threshold=0.8, alpha=0.5, beta=0.5)
for (query_id, hits) in zip(arena.ids, results):
    print(query_id, "->", ", ".join(("%s %.2f" % hit) for hit in  hits.get_ids_and_scores()))
Parameters:
  • arena (a chemfp.arena.FingerprintArena) – the set of fingerprints
  • k (positive integer) – the number of nearest neighbors to find.
  • threshold (float between 0.0 and 1.0, inclusive) – The minimum score threshold.
  • alpha (a value between 0.0 and 100.0, inclusive) – the Tversky alpha value
  • beta (a value between 0.0 and 100.0, inclusive) – the Tversky beta value
  • include_lower_triangle (boolean) – if False, compute only the upper triangle, otherwise use symmetry to compute the full matrix
  • batch_size (integer) – the number of rows to process before checking for a ^C
Returns:

a chemfp.search.SearchResults

chemfp.search.contains_fp(query_fp, target_arena)

Find the target fingerprints which contain the query fingerprint bits as a subset

A target fingerprint contains a query fingerprint if all of the on bits of the query fingerprint are also on bits of the target fingerprint. This function returns a chemfp.search.SearchResult containing all of the target fingerprints in target_arena that contain the query_fp.

The SearchResult scores are all 0.0.

There is currently no direct way to limit the arena search range. Instead create a subarena by using Python’s slice notation on the arena then search the subarena.

Parameters:
Returns:

a SearchResult instance

chemfp.search.contains_arena(query_arena, target_arena)

Find the target fingerprints which contain the query fingerprints as a subset

A target fingerprint contains a query fingerprint if all of the on bits of the query fingerprint are also on bits of the target fingerprint. This function returns a chemfp.search.SearchResults where SearchResults[i] contains all of the target fingerprints in target_arena that contain the fingerprint for entry query_arena [i].

The SearchResult scores are all 0.0.

There is currently no direct way to limit the arena search range, though you can create and search a subarena by using Python’s slice notation.

Parameters:
Returns:

a chemfp.search.SearchResults instance, of the same size as query_arena