chython 2.19__cp313-cp313-win_amd64.whl

chython/algorithms/fingerprints/linear.py

@@ -0,0 +1,202 @@
+# -*- coding: utf-8 -*-
+#
+#  Copyright 2021-2026 Ramil Nugmanov <nougmanoff@protonmail.com>
+#  Copyright 2021 Aleksandr Sizov <murkyrussian@gmail.com>
+#  Copyright 2023 Timur Gimadiev <timur.gimadiev@gmail.com>
+#  This file is part of chython.
+#
+#  chython is free software; you can redistribute it and/or modify
+#  it under the terms of the GNU Lesser General Public License as published by
+#  the Free Software Foundation; either version 3 of the License, or
+#  (at your option) any later version.
+#
+#  This program is distributed in the hope that it will be useful,
+#  but WITHOUT ANY WARRANTY; without even the implied warranty of
+#  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+#  GNU Lesser General Public License for more details.
+#
+#  You should have received a copy of the GNU Lesser General Public License
+#  along with this program; if not, see <https://www.gnu.org/licenses/>.
+#
+from collections import defaultdict, deque
+from math import log2
+from numpy import ndarray, uint8, zeros
+
+
+class LinearFingerprint:
+    __slots__ = ()
+    """
+    Linear fragments fingerprints.
+    Transform structures into fingerprints based on linear fragments descriptors.
+    Also count of fragments takes into account by activating multiple bits,
+    but less or equal to `number_bit_pairs`.To take into account
+    all repeating fragments put 0 as a value of `number_bit_pairs` parameter.
+
+    For example `CC` fragment found 4 times and `number_bit_pairs` is set to 3.
+    In this case will be activated 3 bits: for count 1, for count 2 and for count 3.
+    This gives intersection in bits with another structure with only 2 `CC` fragments.
+    """
+
+    def linear_fingerprint(self, min_radius: int = 1, max_radius: int = 4,
+                           length: int = 1024, number_active_bits: int = 2,
+                           number_bit_pairs: int = 4) -> ndarray:
+        """
+        Transform structures into array of binary features.
+
+        :param min_radius: minimal length of fragments
+        :param max_radius: maximum length of fragments
+        :param length: bit string's length. Should be power of 2
+        :param number_active_bits: number of active bits for each hashed tuple
+        :param number_bit_pairs: describe how much repeating fragments we can count in hashable fingerprint (if
+            number of fragment in molecule greater or equal this number, we will activate only this number of
+            fragments). To take into account all repeating fragments put 0 as a value.
+
+        :return: array(n_features)
+        """
+        bits = self.linear_bit_set(min_radius, max_radius, length, number_active_bits,
+                                   number_bit_pairs)
+        fingerprints = zeros(length, dtype=uint8)
+        fingerprints[list(bits)] = 1
+        return fingerprints
+
+    def linear_bit_set(self, min_radius: int = 1, max_radius: int = 4, length: int = 1024, number_active_bits: int = 2,
+                       number_bit_pairs: int = 4) -> set[int]:
+        """
+        Transform structure into set of indexes of True-valued features.
+
+        :param min_radius: minimal length of fragments
+        :param max_radius: maximum length of fragments
+        :param length: bit string's length. Should be power of 2
+        :param number_active_bits: number of active bits for each hashed tuple
+        :param number_bit_pairs: describe how much repeating fragments we can count in hashable fingerprint (if
+            number of fragment in molecule greater or equal this number, we will activate only this number of
+            fragments). To take into account all repeating fragments put 0 as a value.
+        """
+        mask = length - 1
+        log = int(log2(length))
+
+        hashes = self.linear_hash_set(min_radius, max_radius, number_bit_pairs)
+        active_bits = set()
+        for tpl in hashes:
+            active_bits.add(tpl & mask)
+            if number_active_bits == 2:
+                active_bits.add((tpl >> log) & mask)
+            elif number_active_bits > 2:
+                for _ in range(1, number_active_bits):
+                    tpl >>= log  # shift
+                    active_bits.add(tpl & mask)
+        return active_bits
+
+    def linear_hash_set(self, min_radius: int = 1, max_radius: int = 4, number_bit_pairs: int = 4) -> set[int]:
+        """
+        Transform structure into set of integer hashes of fragments with count information.
+
+        :param min_radius: minimal length of fragments
+        :param max_radius: maximum length of fragments
+        :param number_bit_pairs: describe how much repeating fragments we can count in hashable fingerprint (if
+            number of fragment in molecule greater or equal this number, we will activate only this number of
+            fragments). To take into account all repeating fragments put 0 as a value.
+        """
+        if not number_bit_pairs:
+            number_bit_pairs = 999_999_999  # unreachable count
+
+        return {hash((*tpl, cnt)) for tpl, count in
+                self._fragments(min_radius, max_radius).items()
+                for cnt in range(min(len(count), number_bit_pairs))}
+
+    def linear_hash_smiles(self, min_radius: int = 1, max_radius: int = 4,
+                           number_bit_pairs: int = 4) -> dict[int, list[str]]:
+        """
+        Transform structure into dict of integer hashes of fragments with count information and
+            corresponding fragment SMILES.
+
+        :param min_radius: minimal length of fragments
+        :param max_radius: maximum length of fragments
+        :param number_bit_pairs: describe how much repeating fragments we can count in hashable fingerprint (if
+            number of fragment in molecule greater or equal this number, we will activate only this number of
+            fragments). To take into account all repeating fragments put 0 as a value.
+        """
+        if not number_bit_pairs:
+            number_bit_pairs = 999_999_999  # unreachable count
+
+        out = defaultdict(set)
+        for frg, chains in self._fragments(min_radius, max_radius).items():
+            chain = chains[0]
+            smiles = [self._format_atom(chain[0], None, stereo=False)]
+            for x, y in zip(chain, chain[1:]):
+                smiles.append(self._format_bond(x, y, None, stereo=False, aromatic=False))
+                smiles.append(self._format_atom(y, None, stereo=False))
+            smiles = ''.join(smiles)
+
+            for cnt in range(min(len(chains), number_bit_pairs)):
+                out[hash((*frg, cnt))].add(smiles)  # collisions possible
+        return {k: list(v) for k, v in out.items()}
+
+    def linear_smiles_hash(self, min_radius: int = 1, max_radius: int = 4,
+                           number_bit_pairs: int = 4) -> dict[str, list[int]]:
+        """
+        Transform structure into dict of fragment SMILES and list of corresponding integer hashes of fragments.
+
+        :param min_radius: minimal length of fragments
+        :param max_radius: maximum length of fragments
+        :param number_bit_pairs: describe how much repeating fragments we can count in hashable fingerprint (if
+            number of fragment in molecule greater or equal this number, we will activate only this number of
+            fragments). To take into account all repeating fragments put 0 as a value.
+        """
+        out = defaultdict(list)
+        for k, sl in self.linear_hash_smiles(min_radius, max_radius, number_bit_pairs).items():
+            for s in sl:
+                out[s].append(k)
+        return dict(out)
+
+    def _chains(self, min_radius: int = 1, max_radius: int = 4) -> set[tuple[int, ...]]:
+        queue: deque[tuple[int, ...]]  # typing
+        atoms = self._atoms
+        bonds = self._bonds
+
+        if min_radius == 1:
+            arr = {(x,) for x in atoms}
+            if max_radius == 1:  # special case
+                return arr
+            else:
+                queue = deque(arr)
+        else:
+            arr = set()
+            queue = deque((x,) for x in atoms)
+
+        while queue:
+            now = queue.popleft()
+            var = [now + (x,) for x in bonds[now[-1]] if x not in now]
+            if var:
+                if len(var[0]) < max_radius:
+                    queue.extend(var)
+                if len(var[0]) >= min_radius:
+                    for frag in var:
+                        rev = frag[::-1]
+                        arr.add(frag if frag > rev else rev)
+        return arr
+
+    def _fragments(self, min_radius: int = 1, max_radius: int = 4) -> dict[tuple[int, ...], list[tuple[int, ...]]]:
+        atoms = self._atom_identifiers
+        bonds = self._bonds
+        out = defaultdict(list)
+
+        for frag in self._chains(min_radius, max_radius):
+            var = [atoms[frag[0]]]
+            for x, y in zip(frag, frag[1:]):
+                var.append(int(bonds[x][y]))
+                var.append(atoms[y])
+            var = tuple(var)
+            rev_var = var[::-1]
+            if var > rev_var:
+                out[var].append(frag)
+            else:
+                out[rev_var].append(frag[::-1])
+        return dict(out)
+
+    @property
+    def _atom_identifiers(self) -> dict[int, int]:
+        raise NotImplementedError
+
+
+__all__ = ['LinearFingerprint']

chython/algorithms/fingerprints/morgan.py

@@ -0,0 +1,132 @@
+# -*- coding: utf-8 -*-
+#
+#  Copyright 2021-2026 Ramil Nugmanov <nougmanoff@protonmail.com>
+#  Copyright 2023 Timur Gimadiev <timur.gimadiev@gmail.com>
+#  Copyright 2021 Aleksandr Sizov <murkyrussian@gmail.com>
+#  This file is part of chython.
+#
+#  chython is free software; you can redistribute it and/or modify
+#  it under the terms of the GNU Lesser General Public License as published by
+#  the Free Software Foundation; either version 3 of the License, or
+#  (at your option) any later version.
+#
+#  This program is distributed in the hope that it will be useful,
+#  but WITHOUT ANY WARRANTY; without even the implied warranty of
+#  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+#  GNU Lesser General Public License for more details.
+#
+#  You should have received a copy of the GNU Lesser General Public License
+#  along with this program; if not, see <https://www.gnu.org/licenses/>.
+#
+from collections import defaultdict
+from math import log2
+from numpy import ndarray, uint8, zeros
+
+
+class MorganFingerprint:
+    __slots__ = ()
+
+    def morgan_fingerprint(self, min_radius: int = 1, max_radius: int = 4,
+                           length: int = 1024, number_active_bits: int = 2) -> ndarray:
+        """
+        Transform structures into array of binary features.
+        Morgan fingerprints. Similar to RDkit implementation.
+
+        :param min_radius: minimal radius of EC
+        :param max_radius: maximum radius of EC
+        :param length: bit string's length. Should be power of 2
+        :param number_active_bits: number of active bits for each hashed tuple
+
+        :return: array(n_features)
+        """
+        bits = self.morgan_bit_set(min_radius, max_radius, length, number_active_bits)
+        fingerprints = zeros(length, dtype=uint8)
+        fingerprints[list(bits)] = 1
+        return fingerprints
+
+    def morgan_bit_set(self, min_radius: int = 1, max_radius: int = 4,
+                       length: int = 1024, number_active_bits: int = 2) -> set[int]:
+        """
+        Transform structures into set of indexes of True-valued features.
+
+        :param min_radius: minimal radius of EC
+        :param max_radius: maximum radius of EC
+        :param length: bit string's length. Should be power of 2
+        :param number_active_bits: number of active bits for each hashed tuple
+        """
+        mask = length - 1
+        log = int(log2(length))
+
+        active_bits = set()
+        for tpl in self.morgan_hash_set(min_radius, max_radius):
+            active_bits.add(tpl & mask)
+            if number_active_bits == 2:
+                active_bits.add(tpl >> log & mask)
+            elif number_active_bits > 2:
+                for _ in range(1, number_active_bits):
+                    tpl >>= log
+                    active_bits.add(tpl & mask)
+        return active_bits
+
+    def morgan_hash_set(self, min_radius: int = 1, max_radius: int = 4) -> set[int]:
+        """
+        Transform structures into integer hashes of atoms with EC.
+
+        :param min_radius: minimal radius of EC
+        :param max_radius: maximum radius of EC
+        """
+        return {x for x in self._morgan_hash_dict(min_radius, max_radius) for x in x.values()}
+
+    def morgan_hash_smiles(self, min_radius: int = 1, max_radius: int = 4) -> dict[int, list[str]]:
+        """
+        Transform structures into dictionary of hashes of atoms with EC and corresponding SMILES.
+
+        :param min_radius: minimal radius of EC
+        :param max_radius: maximum radius of EC
+        """
+        smiles_dict = defaultdict(set)
+        for radius, hash_dict in enumerate(self._morgan_hash_dict(min_radius, max_radius), min_radius - 1):
+            for atom, morgan_hash in hash_dict.items():
+                smiles_dict[morgan_hash].add(format(self.augmented_substructure((atom,), deep=radius), 'A'))
+        return {k: list(v) for k, v in smiles_dict.items()}
+
+    def morgan_smiles_hash(self, min_radius: int = 1, max_radius: int = 4) -> dict[str, list[int]]:
+        """
+        Transform structures into dictionary of smiles and corresponding hashes of atoms with EC.
+
+        :param min_radius: minimal radius of EC
+        :param max_radius: maximum radius of EC
+        """
+        out = defaultdict(list)
+        for k, sl in self.morgan_hash_smiles(min_radius, max_radius).items():
+            for s in sl:
+                out[s].append(k)
+        return dict(out)
+
+    def _morgan_hash_dict(self, min_radius: int = 1, max_radius: int = 4) -> list[dict[int, int]]:
+        """
+        Transform structures into integer hashes of atoms with EC.
+        Returns list of atom-hash pairs for different radii.
+
+        :param min_radius: minimal radius of EC
+        :param max_radius: maximum radius of EC
+        """
+        assert min_radius >= 1, 'min_radius should be positive'
+        assert max_radius >= min_radius, 'max_radius should be greater or equal to min_radius'
+        identifiers = self._atom_identifiers
+
+        bonds = self._bonds
+        out = [identifiers]
+        for _ in range(1, max_radius):
+            identifiers = {idx: hash((tpl, *(x for x in sorted((int(b), identifiers[ngb])
+                                                               for ngb, b in bonds[idx].items()) for x in x)))
+                           for idx, tpl in identifiers.items()}
+            out.append(identifiers)
+        return out[-(max_radius - min_radius + 1):]  # slice [min, max] radii range
+
+    @property
+    def _atom_identifiers(self) -> dict[int, int]:
+        raise NotImplementedError
+
+
+__all__ = ['MorganFingerprint']

chython/algorithms/fingerprints/test/__init__.py

@@ -0,0 +1,18 @@
+# -*- coding: utf-8 -*-
+#
+#  Copyright 2025, 2026 Ramil Nugmanov <nougmanoff@protonmail.com>
+#  This file is part of chython.
+#
+#  chython is free software; you can redistribute it and/or modify
+#  it under the terms of the GNU Lesser General Public License as published by
+#  the Free Software Foundation; either version 3 of the License, or
+#  (at your option) any later version.
+#
+#  This program is distributed in the hope that it will be useful,
+#  but WITHOUT ANY WARRANTY; without even the implied warranty of
+#  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+#  GNU Lesser General Public License for more details.
+#
+#  You should have received a copy of the GNU Lesser General Public License
+#  along with this program; if not, see <https://www.gnu.org/licenses/>.
+#

chython/algorithms/fingerprints/test/test_linear.py

@@ -0,0 +1,127 @@
+# -*- coding: utf-8 -*-
+#
+#  Copyright 2025, 2026 Ramil Nugmanov <nougmanoff@protonmail.com>
+#  Copyright 2025 Tagir Akhmetshin <tagirshin@gmail.com>
+#  This file is part of chython.
+#
+#  chython is free software; you can redistribute it and/or modify
+#  it under the terms of the GNU Lesser General Public License as published by
+#  the Free Software Foundation; either version 3 of the License, or
+#  (at your option) any later version.
+#
+#  This program is distributed in the hope that it will be useful,
+#  but WITHOUT ANY WARRANTY; without even the implied warranty of
+#  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+#  GNU Lesser General Public License for more details.
+#
+#  You should have received a copy of the GNU Lesser General Public License
+#  along with this program; if not, see <https://www.gnu.org/licenses/>.
+#
+import numpy as np
+from chython import smiles
+from pytest import mark, raises
+
+
+def test_basic():
+    # Test basic fingerprint generation
+    mol = smiles('CCO')
+    fp = mol.linear_fingerprint(min_radius=1, max_radius=2, length=1024)
+
+    # Test array properties
+    assert isinstance(fp, np.ndarray)
+    assert fp.dtype == np.uint8
+    assert fp.shape == (1024,)
+
+    # Test binary nature
+    assert set(np.unique(fp)).issubset({0, 1})
+
+    # Test different lengths
+    fp = mol.linear_fingerprint(length=2048)
+    assert isinstance(fp, np.ndarray)
+    assert fp.dtype == np.uint8
+    assert fp.shape == (2048,)
+
+
+@mark.parametrize('radius', [(0, 1), (1, 0), (-1, 2), (2, 1)])
+def test_invalid_radius(radius):
+    mol = smiles('CCO')
+    min_r, max_r = radius
+    with raises(AssertionError):
+        mol.morgan_fingerprint(min_radius=min_r, max_radius=max_r)
+
+
+def test_consistency():
+    # Test that fingerprints are consistent for the same molecule
+    fp1 = smiles('CCO').linear_fingerprint()
+    fp2 = smiles('OCC').linear_fingerprint()
+
+    # Test exact equality of arrays
+    assert np.array_equal(fp1, fp2)
+
+    # Test different molecules give different fingerprints)
+    fp3 = smiles('CCC').linear_fingerprint()
+    assert not np.array_equal(fp1, fp3)
+
+
+def test_parameters():
+    mol = smiles('CCO')
+
+    # Test different radius parameters
+    fp1 = mol.linear_fingerprint(min_radius=1, max_radius=2)
+    fp2 = mol.linear_fingerprint(min_radius=1, max_radius=3)
+    assert not np.array_equal(fp1, fp2)
+    assert np.array_equal(fp1 & fp2, fp1)
+
+    # Test number of active bits
+    fp3 = mol.linear_fingerprint(number_active_bits=2)
+    fp4 = mol.linear_fingerprint(number_active_bits=3)
+    assert not np.array_equal(fp3, fp4)
+    assert np.array_equal(fp3 & fp4, fp3)
+
+
+def test_bit_pairs():
+    # Test the number_bit_pairs parameter
+    mol = smiles('CCCCCCCCCCCCCCCCCCCCCCCC')  # molecule with multiple similar fragments
+
+    # Compare different number_bit_pairs settings
+    fp1 = mol.linear_fingerprint(number_bit_pairs=2)
+    fp2 = mol.linear_fingerprint(number_bit_pairs=3)
+
+    assert not np.array_equal(fp1, fp2)
+    assert np.array_equal(fp1 & fp2, fp1)
+
+
+def test_edge_cases():
+    fp1 = smiles('C').linear_fingerprint()
+    assert fp1.sum() == 2
+
+    fp1 = smiles('CC').linear_fingerprint()
+    fp2 = smiles('CC.CC').linear_fingerprint()
+    assert not np.array_equal(fp1, fp2)
+    assert np.array_equal(fp1 & fp2, fp1)
+
+
+def test_comparison():
+    # Test fingerprint comparison between similar molecules
+    mol1 = smiles('CCO')
+    mol2 = smiles('CCC')
+    mol3 = smiles('CCCO')
+
+    fp1 = mol1.linear_fingerprint()
+    fp2 = mol2.linear_fingerprint()
+    fp3 = mol3.linear_fingerprint()
+
+    # Calculate Tanimoto similarities
+    def tanimoto(a, b):
+        intersection = np.sum(np.logical_and(a, b))
+        union = np.sum(np.logical_or(a, b))
+        return intersection / union if union > 0 else 0.0
+
+    # Similar molecules should have higher similarity
+    sim12 = tanimoto(fp1, fp2)
+    sim13 = tanimoto(fp1, fp3)
+    sim23 = tanimoto(fp2, fp3)
+
+    assert 0 <= sim12 <= 1
+    assert 0 <= sim13 <= 1
+    assert 0 <= sim23 <= 1

chython/algorithms/fingerprints/test/test_morgan.py

@@ -0,0 +1,79 @@
+# -*- coding: utf-8 -*-
+#
+#  Copyright 2025, 2026 Ramil Nugmanov <nougmanoff@protonmail.com>
+#  Copyright 2025 Tagir Akhmetshin <tagirshin@gmail.com>
+#  This file is part of chython.
+#
+#  chython is free software; you can redistribute it and/or modify
+#  it under the terms of the GNU Lesser General Public License as published by
+#  the Free Software Foundation; either version 3 of the License, or
+#  (at your option) any later version.
+#
+#  This program is distributed in the hope that it will be useful,
+#  but WITHOUT ANY WARRANTY; without even the implied warranty of
+#  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+#  GNU Lesser General Public License for more details.
+#
+#  You should have received a copy of the GNU Lesser General Public License
+#  along with this program; if not, see <https://www.gnu.org/licenses/>.
+#
+from chython import smiles
+import numpy as np
+from pytest import mark, raises
+
+
+def test_basic():
+    # Test basic fingerprint generation
+    mol = smiles('CCO')
+    fp = mol.morgan_fingerprint(min_radius=1, max_radius=2, length=1024)
+
+    # Test array properties
+    assert isinstance(fp, np.ndarray)
+    assert fp.dtype == np.uint8
+    assert fp.shape == (1024,)
+
+    # Test binary nature
+    assert set(np.unique(fp)).issubset({0, 1})
+
+    # Test different lengths
+    fp = mol.morgan_fingerprint(length=2048)
+    assert isinstance(fp, np.ndarray)
+    assert fp.dtype == np.uint8
+    assert fp.shape == (2048,)
+
+
+@mark.parametrize('radius', [(0, 1), (1, 0), (-1, 2), (2, 1)])
+def test_invalid_radius(radius):
+    mol = smiles('CCO')
+    min_r, max_r = radius
+    with raises(AssertionError):
+        mol.morgan_fingerprint(min_radius=min_r, max_radius=max_r)
+
+
+def test_consistency():
+    # Test that fingerprints are consistent for the same molecule
+    fp1 = smiles('CCO').morgan_fingerprint()
+    fp2 = smiles('OCC').morgan_fingerprint()
+
+    # Test exact equality of arrays
+    assert np.array_equal(fp1, fp2)
+
+    # Test different molecules give different fingerprints)
+    fp3 = smiles('CCC').morgan_fingerprint()
+    assert not np.array_equal(fp1, fp3)
+
+
+def test_parameters():
+    mol = smiles('CCO')
+
+    # Test different radius parameters
+    fp1 = mol.morgan_fingerprint(min_radius=1, max_radius=2)
+    fp2 = mol.morgan_fingerprint(min_radius=1, max_radius=3)
+    assert not np.array_equal(fp1, fp2)
+    assert np.array_equal(fp1 & fp2, fp1)
+
+    # Test number of active bits
+    fp3 = mol.morgan_fingerprint(number_active_bits=2)
+    fp4 = mol.morgan_fingerprint(number_active_bits=3)
+    assert not np.array_equal(fp3, fp4)
+    assert np.array_equal(fp3 & fp4, fp3)