Source code for stk.databases.mongo_db.constructed_molecule

"""
Constructed Molecule MongoDB
============================

"""

import itertools
from functools import lru_cache

from stk.serialization import (
    ConstructedMoleculeDejsonizer,
    ConstructedMoleculeJsonizer,
)
from stk.utilities import dedupe

from ..constructed_molecule import ConstructedMoleculeDatabase
from .utilities import HashableDict, get_any_value


[docs]class ConstructedMoleculeMongoDb(ConstructedMoleculeDatabase): """ Uses MongoDB to store and retrieve constructed molecules. See Also -------- :class:`.MoleculeMongoDb` If you need to store and retrieve molecules, which are not :class:`.ConstructedMolecule` instances, use a \ :class:`.MoleculeMongoDb`. Examples -------- *Storing and Retrieving Constructed Molecules* You want to store and retrieve a :class:`.ConstructedMolecule` from the database .. testsetup:: storing-and-retrieving-constructed-molecules import stk # Change the default database used, so that when a developer # runs the doctests locally, their "stk" database is not # contaminated. _test_database = '_stk_doctest_database' _old_init = stk.ConstructedMoleculeMongoDb stk.ConstructedMoleculeMongoDb = lambda mongo_client: ( _old_init( mongo_client=mongo_client, database=_test_database, ) ) # Change the database MongoClient will connect to. import os import pymongo _mongo_client = pymongo.MongoClient _mongodb_uri = os.environ.get( 'MONGODB_URI', 'mongodb://localhost:27017/' ) pymongo.MongoClient = lambda: _mongo_client(_mongodb_uri) .. testcode:: storing-and-retrieving-constructed-molecules import stk import pymongo # Connect to a MongoDB. This example connects to a local # MongoDB, but you can connect to a remote DB too with # MongoClient() - read the documentation for pymongo to see how # to do that. client = pymongo.MongoClient() db = stk.ConstructedMoleculeMongoDb(client) # Create a molecule. polymer = stk.ConstructedMolecule( topology_graph=stk.polymer.Linear( building_blocks=( stk.BuildingBlock('BrCCBr', [stk.BromoFactory()]), ), repeating_unit='A', num_repeating_units=2, ), ) # Place it into the database. db.put(polymer) # Retrieve it from the database. key_maker = stk.InchiKey() retrieved = db.get({ key_maker.get_key_name(): key_maker.get_key(polymer), }) .. testcode:: storing-and-retrieving-constructed-molecules :hide: _smiles = stk.Smiles() assert _smiles.get_key(polymer) == _smiles.get_key(retrieved) .. testcleanup:: storing-and-retrieving-constructed-molecules pymongo.MongoClient().drop_database(_test_database) stk.ConstructedMoleculeMongoDb = _old_init pymongo.MongoClient = _mongo_client Note that the molecule retrieved from that database can have a different atom ordering than the one put into it. So while the molecule will have the same structure, the order of the atoms may be different to the molecule placed into the database. This is because the database gives the molecules a canonical atom ordering, which allows position matrices to be used across different atom id orderings. *Iterating over All Entries in the Database* All entries in a database can be iterated over very simply .. testsetup:: iterating-over-all-entries-in-the-database import stk import pymongo import os # Change the database used, so that when a developer # runs the doctests locally, their "stk" database is not # contaminated. _test_database = '_stk_doctest_database' # Change the database MongoClient will connect to. _mongodb_uri = os.environ.get( 'MONGODB_URI', 'mongodb://localhost:27017/' ) client = pymongo.MongoClient(_mongodb_uri) db = stk.ConstructedMoleculeMongoDb( mongo_client=client, database=_test_database, ) # Create a molecule. polymer = stk.ConstructedMolecule( topology_graph=stk.polymer.Linear( building_blocks=( stk.BuildingBlock('BrCCBr', [stk.BromoFactory()]), ), repeating_unit='A', num_repeating_units=2, ), ) # Place it into the database. db.put(polymer) .. testcode:: iterating-over-all-entries-in-the-database for entry in db.get_all(): # Do something to the entry. print(stk.Smiles().get_key(entry)) .. testoutput:: iterating-over-all-entries-in-the-database :hide: BrCCCCBr .. testcleanup:: iterating-over-all-entries-in-the-database pymongo.MongoClient(_mongodb_uri).drop_database(_test_database) *Using Alternative Keys for Retrieving Molecules* By default, the only molecular key the database stores, is the InChIKey. However, additional keys can be added to the JSON stored in the database by using a different :class:`.ConstructedMoleculeJsonizer` .. testsetup:: using-alternative-keys-for-retrieving-molecules import stk # Change the database used, so that when a developer # runs the doctests locally, their "stk" database is not # contaminated. _test_database = '_stk_doctest_database' _old_init = stk.ConstructedMoleculeMongoDb stk.ConstructedMoleculeMongoDb = ( lambda mongo_client, jsonizer: _old_init( mongo_client=mongo_client, database=_test_database, jsonizer=jsonizer, ) ) # Change the database MongoClient will connect to. import os import pymongo _mongo_client = pymongo.MongoClient _mongodb_uri = os.environ.get( 'MONGODB_URI', 'mongodb://localhost:27017/' ) pymongo.MongoClient = lambda: _mongo_client(_mongodb_uri) .. testcode:: using-alternative-keys-for-retrieving-molecules import stk import pymongo db = stk.ConstructedMoleculeMongoDb( mongo_client=pymongo.MongoClient(), # Store the InChI and the InChIKey of molecules in # the JSON representation. jsonizer=stk.ConstructedMoleculeJsonizer( key_makers=(stk.Inchi(), stk.InchiKey()), ), ) # Create a molecule. polymer = stk.ConstructedMolecule( topology_graph=stk.polymer.Linear( building_blocks=( stk.BuildingBlock('BrCCBr', [stk.BromoFactory()]), ), repeating_unit='A', num_repeating_units=2, ), ) # Places the JSON of the molecule into the database. In this # case, the JSON includes both the InChI and the InChIKey. db.put(polymer) # You can now use the InChI or the InChIKey to retrieve the # molecule from the database. key_maker = stk.Inchi() retrieved = db.get({ key_maker.get_key_name(): key_maker.get_key(polymer), }) .. testcode:: using-alternative-keys-for-retrieving-molecules :hide: _smiles = stk.Smiles() assert _smiles.get_key(polymer) == _smiles.get_key(retrieved) Obviously, most of the time, you won't have the molecule you are trying to retrieve from the database. Maybe you only have the SMILES of the molecule. You can still retrieve it .. testcode:: using-alternative-keys-for-retrieving-molecules import rdkit.Chem.AllChem as rdkit retrieved2 = db.get({ 'InChI': rdkit.MolToInchi(rdkit.MolFromSmiles('BrCCCCBr')), }) .. testcode:: using-alternative-keys-for-retrieving-molecules :hide: assert _smiles.get_key(polymer) == _smiles.get_key(retrieved2) As long as you have the name of the key, and the expected value of the key, you can retrieve your molecule from the database. Note that you can create your own keys and add them to the database .. testcode:: using-alternative-keys-for-retrieving-molecules # Create your own key. This one is called "SMILES" and the # value is the SMILES of the molecule. class Smiles(stk.MoleculeKeyMaker): def __init__(self): return def get_key_name(self): return 'SMILES' def get_key(self, molecule): return rdkit.MolToSmiles(molecule.to_rdkit_mol()) db = stk.ConstructedMoleculeMongoDb( mongo_client=pymongo.MongoClient(), jsonizer=stk.ConstructedMoleculeJsonizer( # Include your own custom key maker in the JSON # representation. key_makers = (stk.Inchi(), stk.InchiKey(), Smiles()), ), ) # Place the JSON of your molecule into the database. In this # case the JSON will include a key called "SMILES" and # the value will be the SMILES of the molecule. db.put(polymer) # You can now find your molecule by using SMILES as the key, def normalize_smiles(smiles): return rdkit.MolToSmiles( mol=rdkit.AddHs(rdkit.MolFromSmiles(smiles)), ) retrieved3 = db.get({'SMILES': normalize_smiles('BrCCCCBr')}) .. testcode:: using-alternative-keys-for-retrieving-molecules :hide: assert _smiles.get_key(polymer) == _smiles.get_key(retrieved3) Often, it is unnecessary to create a whole subclass for a your custom key .. testcode:: using-alternative-keys-for-retrieving-molecules smiles = stk.MoleculeKeyMaker( key_name='SMILES', get_key=lambda molecule: rdkit.MolToSmiles(molecule.to_rdkit_mol()), ) db = stk.ConstructedMoleculeMongoDb( mongo_client=pymongo.MongoClient(), jsonizer=stk.ConstructedMoleculeJsonizer( key_makers=(stk.InchiKey(), smiles), ) ) .. testcode:: using-alternative-keys-for-retrieving-molecules :hide: db.put(polymer) _retrieved4 = db.get({'SMILES': smiles.get_key(polymer)}) assert _smiles.get_key(polymer) == _smiles.get_key(_retrieved4) .. testcleanup:: using-alternative-keys-for-retrieving-molecules pymongo.MongoClient().drop_database(_test_database) stk.ConstructedMoleculeMongoDb = _old_init pymongo.MongoClient = _mongo_client Note that the key you use to get the molecule back from the database should be unique. In other words, there should always just be one molecule which has that key in the database. Using a key that is matched by multiple molecules will likely cause your data to be jumbled. For example, you might return the atoms of one of the molecules matched by the key but holding the position matrix of the second molecule matched by the key. """
[docs] def __init__( self, mongo_client, database='stk', molecule_collection='molecules', constructed_molecule_collection='constructed_molecules', position_matrix_collection='position_matrices', building_block_position_matrix_collection=( 'building_block_position_matrices' ), jsonizer=ConstructedMoleculeJsonizer(), dejsonizer=ConstructedMoleculeDejsonizer(), put_lru_cache_size=128, get_lru_cache_size=128, indices=('InChIKey', ), ): """ Initialize a :class:`.ConstructedMoleculeMongoDb`. Parameters ---------- mongo_client : :class:`pymongo.MongoClient` The database client. database : :class:`str` The name of the database to use. molecule_collection : :class:`str` The name of the collection which stores molecular information. constructed_molecule_collection : :class:`str` The name of the collection which stored constructed molecule information, that does not belong in the `molecule_collection`. position_matrix_collection : :class:`str` The name of the collection which stores the position matrices of the molecules put into and retrieved from the database. building_block_position_matrix_collection : :class:`str` The name of the collection, which stores the position matrices of the building blocks of the constructed molecules put into and retrieved from the database. jsonizer : :class:`.ConstructedMoleculeJsonizer` Used to create the JSON representations of molecules stored in the database. dejsonizer : :class:`.ConstructedMoleculeDejsonizer` Used to create :class:`.Molecule` instances from their JSON representations. put_lru_cache_size : :class:`int`, optional A RAM-based least recently used cache is used to avoid writing to the database repeatedly. This sets the number of values which fit into the LRU cache. If ``None``, the cache size will be unlimited. get_lru_cache_size : :class:`int`, optional A RAM-based least recently used cache is used to avoid reading from the database repeatedly. This sets the number of values which fit into the LRU cache. If ``None``, the cache size will be unlimited. indices : :class:`tuple` of :class:`str`, optional The names of molecule keys, on which an index should be created, in order to minimize lookup time. """ database = mongo_client[database] self._molecules = database[molecule_collection] self._constructed_molecules = database[ constructed_molecule_collection ] self._position_matrices = database[position_matrix_collection] self._building_block_position_matrices = database[ building_block_position_matrix_collection ] self._jsonizer = jsonizer self._dejsonizer = dejsonizer self._get = lru_cache(maxsize=get_lru_cache_size)(self._get) self._put = lru_cache(maxsize=put_lru_cache_size)(self._put) for index in indices: # Do not create the same index twice. if f'{index}_1' not in self._molecules.index_information(): self._molecules.create_index(index) if ( f'{index}_1' not in self._constructed_molecules.index_information() ): self._constructed_molecules.create_index(index) if ( f'{index}_1' not in self._position_matrices.index_information() ): self._position_matrices.create_index(index) if ( f'{index}_1' not in self._building_block_position_matrices .index_information() ): self._building_block_position_matrices.create_index( index, )
[docs] def put(self, molecule): molecule = molecule.with_canonical_atom_ordering() json = self._jsonizer.to_json(molecule) # lru_cache requires that the parameters to the cached function # are hashable objects. json['matrix']['m'] = tuple( tuple(row) for row in json['matrix']['m'] ) json['matrix'] = HashableDict(json['matrix']) json['molecule'] = HashableDict(json['molecule']) json['constructedMolecule'] = HashableDict( json['constructedMolecule'] ) json['constructedMolecule']['BB'] = tuple(map( HashableDict, json['constructedMolecule']['BB'], )) def make_hashable(json): json['matrix']['m'] = tuple( tuple(row) for row in json['matrix']['m'] ) json['matrix'] = HashableDict(json['matrix']) json['molecule'] = HashableDict(json['molecule']) return HashableDict(json) json['buildingBlocks'] = tuple(map( make_hashable, json['buildingBlocks'], )) return self._put(HashableDict(json))
@staticmethod def _get_query(json): keys = dict(json['matrix']) keys.pop('m') query = {'$or': []} for key, value in keys.items(): query['$or'].append({key: value}) return query def _put(self, json): query = self._get_query(json) self._molecules.update_many( filter=query, update={ '$set': json['molecule'], }, upsert=True, ) self._position_matrices.update_many( filter=query, update={ '$set': json['matrix'], }, upsert=True, ) self._add_building_block_keys_from_database( query=query, building_block_keys=json['constructedMolecule']['BB'], ) self._constructed_molecules.update_many( filter=query, update={ '$set': json['constructedMolecule'], }, upsert=True, ) for building_block_json in json['buildingBlocks']: building_block_query = self._get_query(building_block_json) self._molecules.update_many( filter=building_block_query, update={ '$set': building_block_json['molecule'], }, upsert=True, ) self._building_block_position_matrices.update_many( filter=building_block_query, update={ '$set': building_block_json['matrix'], }, upsert=True, ) def _add_building_block_keys_from_database( self, query, building_block_keys, ): """ Add previously deposited keys to `building_block_keys`. Checks the constructed molecule collection to find all constructed molecule entries which match `query`. All matches should merely be duplicate entries for the same constructed molecule. Each entry for the constructed molecule will have a :class:`list` of building blocks, which were used to construct the constructed molecule. A building block is represented in this :class:`list` through a :class:`dict`, which maps the name of a molecular key (like "SMILES" or "InChIKey") to the appropriate value for that building block. The database may have multiple different dictionaries for the same building block, because each dictionary may hold different molecular keys. These differing dictionaries will be spread across the constructed molecule entries. The various different dictionaries, which all represent the same building block, are merged by this method. The merged dictionary is the one held in `building_block_keys`, which is updated in-place. This means that when `building_block_keys` is used to replace an entry in the database, it does not remove any building block keys already there. Parameters ---------- query : :class:`dict` A query which matches entries, corresponding to a single constructed molecule. building_block_keys : :class:`list` of :class:`dict` Each :class:`dict` represents a building block of the constructed molecule matched by `query`. The :class:`dict` holds the name of a molecular key and its value for that particular building block. Key-value pairs for building block molecular keys already found in the database are added to the dictionaries by this method. Returns ------- None : :class:`NoneType` """ database_building_block_keys = ( molecule_entry['BB'] for molecule_entry in self._constructed_molecules.find(query) ) for entry_building_block_keys in database_building_block_keys: for keys1, keys2 in zip( building_block_keys, entry_building_block_keys, ): keys1.update(keys2)
[docs] def get(self, key): # lru_cache requires that the parameters to the cached function # are hashable objects. return self._get(HashableDict(key))
def _get(self, key): """ Get the molecule with `key` from the cache. Parameters ---------- key : :class:`.HashableDict` The key of a molecule, which is to be returned from the database. Returns ------- :class:`.Molecule` The molecule held in the database under `key`. """ molecule_json = self._molecules.find_one(key) if molecule_json is None: raise KeyError( 'No molecule found in the database with a key of: ' f'{key}' ) constructed_molecule_json = ( self._constructed_molecules.find_one(key) ) if constructed_molecule_json is None: raise KeyError( 'No molecule found in the database with a key of: ' f'{key}' ) position_matrix = self._position_matrices.find_one(key) if position_matrix is None: raise KeyError( 'No position matrix found in the database with a key ' f'of: {key}' ) return self._dejsonizer.from_json( json={ 'molecule': molecule_json, 'constructedMolecule': constructed_molecule_json, 'matrix': position_matrix, 'buildingBlocks': tuple(map( self._get_building_block, constructed_molecule_json['BB'], )) }, ) def _get_building_block(self, key): return { 'molecule': self._molecules.find_one(key), 'matrix': self._building_block_position_matrices.find_one(key), }
[docs] def get_all(self): # Get all potential indices. indices = itertools.chain( self._position_matrices.index_information().values(), self._molecules.index_information().values(), self._constructed_molecules.index_information().values(), ) keys = tuple(dedupe( index['key'][0][0] for index in indices # Ignore "_id" index which is unique in a collection and # cannot be used to match molecular data split across # collections. if index['key'][0][0] != '_id' )) query = [ { '$match': { '$or': [ {key: {'$exists': True}} for key in keys ], }, }, ] query.extend( { '$lookup': { 'from': self._position_matrices.name, 'let': { 'molecule_key': f'${key}', }, 'as': f'posmat_{key}', 'pipeline': [ { '$match': { key: {'$ne': None}, }, }, { '$match': { '$expr': { '$eq': [ f'${key}', '$$molecule_key', ], }, }, }, ], }, } for key in keys ) query.extend( { '$lookup': { 'from': self._molecules.name, 'let': { 'molecule_key': f'${key}', }, 'as': f'mol_{key}', 'pipeline': [ { '$match': { key: {'$ne': None}, }, }, { '$match': { '$expr': { '$eq': [ f'${key}', '$$molecule_key', ], }, }, }, ], }, } for key in keys ) query.append( { '$match': { '$expr': { '$or': [ { '$gt': [ {'$size': f'$posmat_{key}'}, 0 ], } for key in keys ], }, }, }, ) query.append( { '$match': { '$expr': { '$or': [ { '$gt': [ {'$size': f'$mol_{key}'}, 0 ], } for key in keys ], }, }, }, ) cursor = self._constructed_molecules.aggregate(query) for entry in cursor: molecule_document = get_any_value( mapping=entry, keys=(f'mol_{key}' for key in keys), ) position_matrix_document = get_any_value( mapping=entry, keys=(f'posmat_{key}' for key in keys), ) if ( molecule_document is not None and position_matrix_document is not None ): yield self._dejsonizer.from_json({ 'molecule': molecule_document, 'constructedMolecule': entry, 'matrix': position_matrix_document, 'buildingBlocks': tuple(map( self._get_building_block, entry['BB'], )), })