biotite 1.6.0__cp314-cp314-win_amd64.whl

biotite/structure/charges.pyx

@@ -0,0 +1,521 @@
+# This source code is part of the Biotite package and is distributed
+# under the 3-Clause BSD License. Please see 'LICENSE.rst' for further
+# information.
+
+"""
+This module provides one function for the computation of the partial
+charges of the individual atoms of a given AtomArray according to the
+PEOE algorithm of Gasteiger-Marsili.
+"""
+
+__name__ = "biotite.structure"
+__author__ = "Jacob Marcel Anter, Patrick Kunzmann"
+__all__ = ["partial_charges"]
+
+cimport numpy as np
+from libc.math cimport isnan
+
+import warnings
+import numpy as np
+from .bonds import BondType
+from .filter import filter_heavy
+
+
+ctypedef np.float32_t float32
+ctypedef np.uint32_t uint32
+ctypedef np.int8_t int8
+ctypedef np.int64_t int64
+
+
+cdef float32 NAN = np.nan
+# Conversion of Enum into int for faster comparison in _get_parameters()
+cdef uint32 ANY = BondType.ANY
+
+
+# Creating two dictionaries to retrieve parameters for electronegativity
+# computation from
+# The first dictionary uses BondTypes as keys, whereas the second uses
+# the amounts of binding partners as keys
+# First level of dictionaries represents the atom name and the second
+# level represents the hybridisation state
+# Depending on whether the bond type is unequal to zero or not,
+# identification of the hybridisation state is either performed via the
+# first dictionary, i. e. the bond type (primary way of identification
+# since erroneous results are excluded) or via the second dictionary,
+# i. e. amount of binding partners (erroneous if a binding partner is
+# lost or added whilst the hybridisation remains unchanged as it is the
+# case with acids or bases, e. g. the hydroxyl group in a carboxyl
+# group: When the proton involved in the hydroxyl group is donated, the
+# amount of binding partners of the remaining oxygen is one; this would
+# erroneously lead to an identification of the hybridisation state as
+# sp2 although it is still sp3)
+cdef dict EN_PARAM_BTYPE = {
+    "H": {
+        int(BondType.SINGLE):   (7.17, 6.24, -0.56)
+    },
+
+    "C": {
+        int(BondType.SINGLE):   (7.98, 9.18, 1.88),
+        int(BondType.DOUBLE):   (8.79, 9.18, 1.88),
+        int(BondType.TRIPLE):   (10.39, 9.45, 0.73),
+        int(BondType.AROMATIC_SINGLE): (7.98, 9.18, 1.88),
+        int(BondType.AROMATIC_DOUBLE): (8.79, 9.18, 1.88)
+    },
+
+    "N": {
+        int(BondType.SINGLE):   (11.54, 10.82, 1.36),
+        int(BondType.DOUBLE):   (12.87, 11.15, 0.85),
+        int(BondType.TRIPLE):   (15.68, 11.7, -0.27),
+        int(BondType.AROMATIC_SINGLE): (11.54, 10.82, 1.36),
+        int(BondType.AROMATIC_DOUBLE): (12.87, 11.15, 0.85)
+    },
+
+    # As oxygen and sulfur are exclusively involved in aromatic
+    # systems having single bonds on either side, the values for a
+    # sp3 hybridisation are taken for BondType.AROMATIC_SINGLE
+    "O": {
+        int(BondType.SINGLE):   (14.18, 12.92, 1.39),
+        int(BondType.DOUBLE):   (17.07, 13.79, 0.47),
+        int(BondType.AROMATIC_SINGLE): (14.18, 12.92, 1.39)
+    },
+
+    "S": {
+        int(BondType.SINGLE):   (10.14, 9.13, 1.38),
+        int(BondType.AROMATIC_SINGLE): (10.14, 9.13, 1.38)
+    },
+
+    "F": {
+        int(BondType.SINGLE):   (14.66, 13.85, 2.31)
+    },
+
+    "Cl": {
+        int(BondType.SINGLE):   (11.00, 9.69, 1.35)
+    },
+
+    "Br": {
+        int(BondType.SINGLE):   (10.08, 8.47, 1.16)
+    },
+
+    "I": {
+        int(BondType.SINGLE):   (9.90, 7.96, 0.96)
+    }
+}
+
+cdef dict EN_PARAM_BPARTNERS = {
+    "H": {
+        1: (7.17, 6.24, -0.56)
+    },
+
+    "C": {
+        4: (7.98, 9.18, 1.88),
+        3: (8.79, 9.18, 1.88),
+        2: (10.39, 9.45, 0.73)
+    },
+
+    "N": {
+        # Considering protonated, e. g. in terminal
+        # amino group (4 binding partners), as well
+        # as unprotonated nitrogen (3 binding partners)
+        4: (11.54, 10.82, 1.36),
+        3: (11.54, 10.82, 1.36),
+        2: (12.87, 11.15, 0.85),
+        1: (15.68, 11.7, -0.27)
+    },
+
+    "O": {
+        2: (14.18, 12.92, 1.39),
+        1: (17.07, 13.79, 0.47)
+    },
+
+    "S": {
+        2: (10.14, 9.13, 1.38)
+    },
+
+    "F": {
+        1: (14.66, 13.85, 2.31)
+    },
+
+    "Cl": {
+        1: (11.00, 9.69, 1.35)
+    },
+
+    "Br": {
+        1: (10.08, 8.47, 1.16)
+    },
+
+    "I": {
+        1: (9.90, 7.96, 0.96)
+    }
+}
+
+
+# Defining constant for the special case of the electronegativity of
+# positively charged hydrogen (value given in electronvolt, as all
+# electronegativity values)
+EN_POS_HYDROGEN = 20.02
+
+
+def _get_parameters(elements, bond_types, amount_of_binding_partners):
+    """
+    Gather the parameters required for electronegativity computation of
+    all atoms comprised in the input array `elements`.
+
+    By doing so, the function accesses the nested dictionary
+    ``EN_PARAMETERS``. The values originate from a publication of Johann
+    Gasteiger and Mario Marsili.
+
+    Parameters
+    ----------
+    elements: ndarray, dtype=str
+        The array comprising the elements which to retrieve the
+        parameters for.
+    bond_types: ndarray, dtype=int
+        The array containing information about the highest bond type of
+        the respective atom (except for the aromatic bond types).
+    amount_of_binding_partners: ndarray, dtype=int
+        The array containing information about the amount of binding
+        partners of the respective atom.
+
+    Returns
+    -------
+    parameters: NumPy array, dtype=float, shape=(n,3)
+        The array containing all three parameters required for the
+        computation of the electronegativities of all atoms comprised
+        in the `elements` array.
+    """
+    cdef int i
+    cdef float32 a, b, c
+    cdef int8[:] bond_types_v = bond_types
+    cdef int64[:] amount_of_binding_partners_v = amount_of_binding_partners
+
+    parameters = np.zeros((elements.shape[0], 3), dtype=np.float32)
+    cdef float32[:,:] parameters_v = parameters
+
+    has_atom_key_error = False
+    has_valence_key_error = False
+    # Preparing warning in case of KeyError
+    # It is differentiated between atoms that are not parametrized at
+    # all and specific valence states that are parametrized
+    list_of_unparametrized_elements = []
+    unparametrized_valences = []
+    unparam_valence_names = []
+    list_of_atoms_without_specified_btype = []
+
+    for i, element in enumerate(elements):
+        # Considering the special case of ions
+        if amount_of_binding_partners_v[i] == 0:
+            parameters_v[i, :] = np.nan
+            continue
+        if bond_types_v[i] == ANY:
+            list_of_atoms_without_specified_btype.append(str(i))
+            try:
+                a, b, c = EN_PARAM_BPARTNERS[element] \
+                                            [amount_of_binding_partners_v[i]]
+                parameters_v[i, 0] = a
+                parameters_v[i, 1] = b
+                parameters_v[i, 2] = c
+            except KeyError:
+                try:
+                    EN_PARAM_BPARTNERS[element]
+                except KeyError:
+                    list_of_unparametrized_elements.append(element)
+                    has_atom_key_error = True
+                else:
+                    # The warning message printed in case of
+                    # unparametrized valence states contains its main
+                    # information in a table with three columns:
+                    # The first column represents the element, the
+                    # second the amount of binding partners and the
+                    # third the BondType
+                    # The primary way of identifying unparametrized
+                    # valence states that is aimed at is via the
+                    # BondType; if this is possible, the space beneath
+                    # the column representing the amount of binding
+                    # partners is padded with a respective amount the
+                    # '-' (hyphen) character
+                    # If not, the space beneath the column representing
+                    # the BondTypes is padded with a respective amount
+                    # of hyphens
+                    # At either case, an appropriate amount of
+                    # whitespace is added in order to ensure that the
+                    # respective entries appear directly under the
+                    # respective columns
+                    unparam_valence_names.append(element)
+                    unparametrized_valences.append(
+                        str(amount_of_binding_partners_v[i])
+                        +
+                        " " * 31
+                        +
+                        "-" * 10
+                    )
+                    has_valence_key_error = True
+                parameters_v[i, :] = np.nan
+        else:
+            try:
+                a, b, c = EN_PARAM_BTYPE[element][bond_types_v[i]]
+                parameters_v[i, 0] = a
+                parameters_v[i, 1] = b
+                parameters_v[i, 2] = c
+            except KeyError:
+                try:
+                    EN_PARAM_BTYPE[element]
+                except KeyError:
+                    list_of_unparametrized_elements.append(element)
+                    has_atom_key_error = True
+                else:
+                    unparam_valence_names.append(element)
+                    unparametrized_valences.append(
+                        "-" * 27
+                        +
+                        " " * 5
+                        +
+                        str(bond_types_v[i])
+                    )
+                    has_valence_key_error = True
+                parameters_v[i, :] = np.nan
+
+
+    # Error and warning handling
+    if np.all(bond_types == BondType.ANY):
+        warnings.warn(
+            f"Each atom's bond type is 0 (any). Therefore, it is "
+            f"resorted to the amount of binding partners for the "
+            f"identification of the hybridisation state which can lead "
+            f"to erroneous results.",
+            UserWarning
+        )
+    elif np.any(bond_types == BondType.ANY):
+        warnings.warn(
+            f"Some atoms' bond type is unspecified, i. e. the bond "
+            f"type is given as `any`. For these atoms, identification "
+            f"of the hybridisation state is performed via the amount "
+            f"of binding partners which can lead to erroneous results."
+            f"\n\n"
+            f"In detail, these atoms possess the following indices: \n"
+            f"{', '. join(list_of_atoms_without_specified_btype)}.",
+            UserWarning
+        )
+
+    if has_valence_key_error:
+        joined_list = []
+        for i in range(len(unparam_valence_names)):
+            joined_list.append(
+                unparam_valence_names[i].ljust(2, ' ')
+                +
+                " " * 8
+            )
+            joined_list.append(unparametrized_valences[i] + "\n")
+        joined_array = np.reshape(
+            joined_list,
+            (int(len(joined_list) / 2), 2)
+        )
+        joined_array = np.unique(joined_array, axis=0)
+        # Array must be flattened in order ro be able to apply the
+        # 'join' method
+        flattened_joined_array = np.reshape(
+            joined_array, (2*joined_array.shape[0])
+        )
+        warnings.warn(
+            f"Parameters for specific valence states of some atoms "
+            f"are not available. These valence states are: \n"
+            f"Atom:     Amount of binding partners:     Bond type:\n"
+            f"{''.join(flattened_joined_array)}"
+            f"Their electronegativity is given as NaN.",
+            UserWarning
+        )
+
+    if has_atom_key_error:
+        # Using NumPy's 'unique' function to ensure that each atom only
+        # occurs once in the list
+        unique_list = np.unique(list_of_unparametrized_elements)
+        # Considering proper punctuation for the warning string
+        warnings.warn(
+            f"Parameters required for computation of "
+            f"electronegativity aren't available for the following "
+            f"atoms: {', '.join(unique_list)}. "
+            f"Their electronegativity is given as NaN.",
+            UserWarning
+        )
+
+
+    return parameters
+
+
+def partial_charges(atom_array, int iteration_step_num=6, charges=None):
+    """
+    partial_charges(atom_array, iteration_step_num=6, charges=None)
+
+    Compute the partial charge of the individual atoms comprised in a
+    given :class:`AtomArray` depending on their electronegativity.
+
+    This function implements the
+    *partial equalization of orbital electronegativity* (PEOE)
+    algorithm :footcite:`Gasteiger1980`.
+
+    Parameters
+    ----------
+    atom_array: AtomArray, shape=(n,)
+        The :class:`AtomArray` to get the partial charge values for.
+        Must have an associated `BondList`.
+    iteration_step_num: int, optional
+        The number of iteration steps is an optional argument and can be
+        chosen by the user depending on the desired precision of the
+        result. If no value is entered by the user, the default value
+        ``6`` will be used.
+        Gasteiger and Marsili described this number as sufficient.
+    charges: ndarray, dtype=int, optional
+        The array comprising the formal charges of the atoms in the
+        input `atom_array`.
+        If none is given, the ``charge`` annotation category of the
+        input `atom_array` is used.
+        If neither of them is given, the formal charges of all atoms
+        will be arbitrarily set to zero.
+
+    Returns
+    -------
+    charges: ndarray, dtype=float32
+        The partial charge values of the individual atoms in the input
+        `atom_array`.
+
+    Notes
+    -----
+    A :class:`BondList` must be associated to the input
+    :class:`AtomArray`.
+    Otherwise, an error will be raised.
+    Example:
+
+    .. code-block:: python
+
+        atom_array.bonds = struc.connect_via_residue_names(atom_array)
+
+    |
+
+    For the electronegativity of positively charged hydrogen, the
+    value of 20.02 eV is used.
+
+    Also note that the algorithm used in this function does not deliver
+    proper results for expanded pi-electron systems like aromatic rings.
+
+    References
+    ----------
+
+    .. footbibliography::
+
+    Examples
+    --------
+
+    >>> fluoromethane = residue("CF0")
+    >>> print(fluoromethane.atom_name)
+    ['C1' 'F1' 'H1' 'H2' 'H3']
+    >>> print(partial_charges(fluoromethane, iteration_step_num=1))
+    [ 0.115 -0.175  0.020  0.020  0.020]
+    >>> print(partial_charges(fluoromethane, iteration_step_num=6))
+    [ 0.079 -0.253  0.058  0.058  0.058]
+    """
+    if atom_array.bonds is None:
+        raise AttributeError(
+            f"The input AtomArray doesn't possess an associated "
+            f"BondList."
+        )
+
+    if charges is None:
+        try:
+            # Implicitly this creates a copy of the charges
+            charges = atom_array.charge.astype(np.float32)
+        except AttributeError:
+            charges = np.zeros(atom_array.shape[0], dtype=np.float32)
+            warnings.warn(
+                f"A charge array was neither given as optional "
+                f"argument, nor does a charge annotation of the "
+                f"inserted AtomArray exist. Therefore, all atoms' "
+                f"formal charge is assumed to be zero.",
+                UserWarning
+            )
+    cdef float32[:] charges_v = charges
+
+    elements = atom_array.element
+    bonds, types = atom_array.bonds.get_all_bonds()
+    amount_of_binding_partners = np.count_nonzero(bonds != -1, axis=1) \
+                                 .astype(np.int64, copy=False)
+    # The maximum of a given row of the `types` array must be determined
+    # as this value reveals the hybridisation state
+    # An atom's overall BondType is assumed to be ANY as soon as one
+    # BondType.ANY occurs
+    if types.shape[1] == 0:
+        # No bonds between atoms
+        # -> types has shape (n,0)
+        # -> np.amax() raises ValueError
+        # This is the case e. g. if AtomArray exclusively contains ions
+        # In this case partial charges are equal to formal charges
+        return charges
+    bond_types = np.amax(types, axis=1)
+    bond_types[np.any(types == BondType.ANY, axis=1)] = BondType.ANY
+    parameters = _get_parameters(
+        elements, bond_types, amount_of_binding_partners
+    )
+
+    # Computing electronegativity values in case of positive charge
+    # which enter as divisor the equation for charge transfer
+    pos_en_values = np.sum(parameters, axis=1)
+    # Substituting values for hydrogen with the special value
+    pos_en_values[~filter_heavy(atom_array)] = EN_POS_HYDROGEN
+    cdef float32[:] pos_en_values_v = pos_en_values
+
+
+    cdef float32 damping = 1.0
+    cdef float32 divisor
+    cdef float32 charge_transfer
+    cdef uint32[:,:] bonds_v = atom_array.bonds.as_array()
+    cdef int bond_index
+    # Indices to atoms involved in a bonds
+    cdef uint32 i, j
+    cdef float32[:] en_values_v
+
+    for _ in range(iteration_step_num):
+        # In the beginning of each iteration step, the damping factor is
+        # halved in order to guarantee rapid convergence
+        damping *= 0.5
+        # Calculate electronegativity via vectorization:
+        # X = a + bQ + cQ^2
+        charge_factor = np.stack((
+            np.ones(atom_array.array_length()),
+            charges,
+            charges**2
+        ), axis=-1)
+        en_values_v = np.sum(
+            parameters * charge_factor,
+            axis=1, dtype=np.float32
+        )
+        # Iterate over bonds to transfer charges
+        # based on new electronegativity values
+        for bond_index in range(bonds_v.shape[0]):
+            i = bonds_v[bond_index, 0]
+            j = bonds_v[bond_index, 1]
+            # For atoms that are not available in the dictionary,
+            # but which are incorporated into molecules,
+            # the partial charge is set to NaN
+            if isnan(en_values_v[i]) or isnan(en_values_v[j]):
+                # Determining for which atom exactly no parameters are
+                # available is necessary since the other atom, for which
+                # there indeed are parameters, could be involved in
+                # multiple bonds.
+                # Therefore, setting both charges to NaN would falsify
+                # the result.
+                # The case that both atoms are not parametrized must be
+                # considered as well.
+                if isnan(en_values_v[i]):
+                    charges_v[i] = NAN
+                if isnan(en_values_v[j]):
+                    charges_v[j] = NAN
+            else:
+                if en_values_v[j] > en_values_v[i]:
+                    divisor = pos_en_values_v[i]
+                else:
+                    divisor = pos_en_values_v[j]
+                charge_transfer = (
+                    (en_values_v[j] - en_values_v[i]) / divisor
+                ) * damping
+                charges_v[i] += charge_transfer
+                charges_v[j] -= charge_transfer
+
+    return charges