def pair_features(mol, idx1, idx2, max_distance=7):
features = np.zeros((6 + max_distance + 1))
# bond type
bond = mol.GetBondBetweenAtoms(idx1, idx2)
if bond is not None:
bt = bond.GetBondType()
features[:6] = np.array([
bt == Chem.rdchem.BondType.SINGLE, bt
== Chem.rdchem.BondType.DOUBLE, bt == Chem.rdchem.BondType.TRIPLE,
bt == Chem.rdchem.BondType.AROMATIC,
bond.GetIsConjugated(),
bond.IsInRing()
],
dtype=int)
# whether two atoms are in same ring
rings = mol.GetRingInfo().AtomRings()
for ring in rings:
if idx1 in ring and idx2 in ring and idx1 != idx2:
features[6] = 1
# graph distance between two atoms
distance = rdmolops.GetDistanceMatrix(mol)
distance = np.where(distance < max_distance, distance,
max_distance - 1)[idx1][idx2]
distance = to_categorical(distance, num_classes=max_distance)
features[7:] = distance
return features
def get_molecules():
"""
Constructs rdkit mol objects derrived from the .xyz files. Also returns:
- mol ids (unique numerical ids)
- set of molecule level features
- arrays of xyz coordinates
- euclidean distance matrices
- graph distance matrices.
All objects are returned in dictionaries with 'mol_name' as keys.
"""
mols, mol_ids, mol_feats = {}, {}, {}
xyzs, dist_matrices, graph_dist_matrices = {}, {}, {}
print('Create molecules and distance matrices.')
for i in range(C.N_MOLS):
print_progress(i, C.N_MOLS)
filepath = xyz_filepath_list[i]
mol_name = filepath.split('/')[-1][:-4]
mol, xyz, dist_matrix = mol_from_xyz(filepath) #读取XYZ文件获取结构mol和距离矩阵，坐标
mols[mol_name] = mol
xyzs[mol_name] = xyz
dist_matrices[mol_name] = dist_matrix
mol_ids[mol_name] = i # 数据集中分子序号作为分子的id
# make padded graph distance matrix dataframes
n_atoms = len(xyz)
graph_dist_matrix = pd.DataFrame(
np.pad(rdmolops.GetDistanceMatrix(mol),
[(0, 0), (0, C.MAX_N_ATOMS - n_atoms)],
'constant')) #通过ramolops.GetDistanceMatrix获取 图距离矩阵
graph_dist_matrix['molecule_id'] = n_atoms * [
i
] # eg: CH4 5 * [0] = [0, 0, 0, 0, 0] list数据可以为dataframe赋值
graph_dist_matrices[mol_name] = graph_dist_matrix #字典：value: dataframe
# compute molecule level features
adj_matrix = rdmolops.GetAdjacencyMatrix(
mol) #通过ramolops.GetDistanceMatrix获取 图邻接矩阵
atomic_num_list, _, _ = read_xyz_file(
filepath) #读取XYZ文件获取分子中各原子的原子序数和坐标
dists = dist_matrix.ravel()[np.tril(adj_matrix).ravel() ==
1] #通过邻接矩阵的下三角获取与相邻原子之间的距离
mol_feats[mol_name] = pd.Series(
[np.mean(dists),
np.std(dists),
np.mean(atomic_num_list)],
index=mol_feat_columns) #获取与领接原子之间距离均值和标准差、原子序数的均值(分子级特征)
return mols, mol_ids, mol_feats, xyzs, dist_matrices, graph_dist_matrices #返回训练集所有分子结构mol和分子ids,分子级特征,原子坐标,距离矩阵，图距离矩阵
def get_molecules():
"""
Constructs rdkit mol objects derrived from the .xyz files. Also returns:
- mol ids (unique numerical ids)
- set of molecule level features
- arrays of xyz coordinates
- euclidean distance matrices
- graph distance matrices.
All objects are returned in dictionaries with 'mol_name' as keys.
"""
mols, mol_ids, mol_feats = {}, {}, {}
xyzs, dist_matrices, graph_dist_matrices = {}, {}, {}
print('Create molecules and distance matrices.')
for i in range(C.N_MOLS):
print_progress(i, C.N_MOLS)
filepath = xyz_filepath_list[i]
mol_name = filepath.split('/')[-1][:-4]
mol, xyz, dist_matrix = mol_from_xyz(filepath)
mols[mol_name] = mol
xyzs[mol_name] = xyz
dist_matrices[mol_name] = dist_matrix
mol_ids[mol_name] = i
# make padded graph distance matrix dataframes
n_atoms = len(xyz)
graph_dist_matrix = pd.DataFrame(
np.pad(rdmolops.GetDistanceMatrix(mol),
[(0, 0), (0, C.MAX_N_ATOMS - n_atoms)], 'constant'))
graph_dist_matrix['molecule_id'] = n_atoms * [i]
graph_dist_matrices[mol_name] = graph_dist_matrix
# compute molecule level features
adj_matrix = rdmolops.GetAdjacencyMatrix(mol)
atomic_num_list, _, _ = read_xyz_file(filepath)
dists = dist_matrix.ravel()[np.tril(adj_matrix).ravel() == 1]
mol_feats[mol_name] = pd.Series(
[np.mean(dists),
np.std(dists),
np.mean(atomic_num_list)],
index=mol_feat_columns)
return mols, mol_ids, mol_feats, xyzs, dist_matrices, graph_dist_matrices
def mol_to_vec(smifile, shared_fg, voltype, max_path_length, verbose=False):
"""
mol_to_vec function
Allows for calculating a graph-based sum of atomic contributions in "layers"
based off atomic connectivity ranging away from a specified common
functional group.
arguments:
smifile: a file listing SMILES strings, properties can optionally be provided after each string.
shared_fg: a SMILES pattern shared by provided SMILES strings, to be used as a reference.
voltype: type of volume contributions to measure, current options are "crippen" or "mcgowan".
max_path_length: number of layers to include contributions from base functional group.
verbose: print information to terminal.
returns:
Pandas DataFrame containing sum of atomic contributions for each layer requested in format:
["layer0",..."layerN","Structure" (SMILES string),"Property" (optional)]
"""
# computes the atomic contributions to volume/sterics at discrete number of bond lengths away from a particular atom/functional group
mollist, y_val, vec_df, columns = [], [], [], []
[
columns.append(str(col) + '_' + voltype.lower())
for col in range(0, max_path_length)
]
if shared_fg == False:
sys.exit(
"Please specify a common functional group pattern as a SMILES string using the --fg argument. Exiting."
)
for line in open(smifile):
toks = line.split()
mol2vec = []
base_id = None
if len(
toks
) > 1: # expects a smiles string, followed by a property value on each line
# parse structure from input
smi, prop = toks[0:2]
y_val.append(prop)
elif len(toks) == 1:
smi = toks[0]
mollist.append(smi)
mol = Chem.MolFromSmiles(smi)
if mol is None:
print("Warning - Could not parse SMILES for", smi,
"\nEvaluating with looser constrictions.")
mol = Chem.MolFromSmiles(smi, sanitize=False)
mol.UpdatePropertyCache(strict=False)
if mol is None:
print("Parsing Failed. Skipping this structure")
vec_df.append(pd.Series())
continue
mat = rdmolops.GetDistanceMatrix(mol)
# the origin is defined by a particular functional group/atom of interest shared by all structures
patt = Chem.MolFromSmarts(shared_fg)
# if a functional group SMILES pattern is specified, the base atom is expected to be first in the smiles string
fg_atoms = mol.GetSubstructMatch(patt)
if len(fg_atoms) == 0:
print("ERR: Functional group", shared_fg, "not found in molecule:",
smi)
lower_fg = shared_fg.lower()
patt = Chem.MolFromSmarts(lower_fg)
fg_atoms = mol.GetSubstructMatch(patt)
if len(fg_atoms) == 0:
print(
"Parsing functional group from this structure failed. Skipping this structure"
)
vec_df.append(pd.Series())
continue
else:
if verbose:
print("Found by parsing functional group as", lower_fg)
base_id = mol.GetSubstructMatch(patt)[0]
dist_from_base = mat[base_id]
# This uses Crippen's atomic contributions to molecular refractivity as volumes
# Atomic contributions to logP are also available ...
if voltype.lower() == 'crippen':
molH = Chem.AddHs(mol)
mrContribs = Crippen.rdMolDescriptors._CalcCrippenContribs(molH)
logps, mrs = zip(*mrContribs)
# condense H atom contributions to attached heavy atom
mr, apolsCondensed = crippenHContribs(molH, mrs)
vols = apolsCondensed
elif voltype.lower() == 'mcgowan':
#grab mcgowan volumes
molH = Chem.AddHs(mol)
vols = mcgowanHContribs(molH)
elif voltype.lower() == 'degree':
vols = degreeContribs(mol)
# this is the radial count up to the max_path_length
for level in range(0, max_path_length):
mol2vec.append(0)
for at, dist in enumerate(dist_from_base):
# This will try to exclude the other atoms in the defined FG apart from the base
if dist == level and at not in fg_atoms:
mol2vec[level] += vols[at]
elif at == base_id and level == 0: # add contributions from base atom at level 0
"""Add contributions from other base atoms here ? or not at all"""
mol2vec[level] += vols[at]
# create the vector from the successive graph levels
vec_df.append(mol2vec)
vec_df = pd.DataFrame(vec_df, columns=columns)
vec_df['Structure'] = np.array(mollist)
if len(y_val) > 0:
vec_df['Property'] = np.array(y_val)
return vec_df
def add_sample(url,features,target,As,sizes,num_molecules):
"""Extracts information from .xyz file and returns the features array,
target array, array of adjacency matrices, array of molecule sizes, and total
number of molecules."""
try:
properties = [];
with open(url,'r') as file:
for row in file:
properties += row.split()
SMILES = properties[-4]
#INCHI = properties[-2]
m = Chem.MolFromSmiles(SMILES)
#m = Chem.AddHs(m) #uncomment to include hydrogen atoms
vertices = m.GetAtoms()
d = len(vertices)
partial_charges=properties[22:(23+5*(d-1)):5]
rdPartialCharges.ComputeGasteigerCharges(m)
# Targets
dipole_moment = float(properties[6])
polarizability = float(properties[7])
h**o = float(properties[8])
lumo = float(properties[9])
gap = float(properties[10])
r2 = float(properties[11])
zpve = float(properties[12])
U0 = float(properties[13])
internal_energy = float(properties[14])
enthalpy = float(properties[15])
free_nrg = float(properties[16])
heat_capacity = float(properties[17])
# This is where you decide which properties you would like to predict
target.append([heat_capacity])
#target.append([dipole_moment,polarizability,h**o,lumo,gap,
# r2,zpve,U0,internal_energy,enthalpy,free_nrg,heat_capacity])
#populate the adjacency matrix and write features
tempA = rdmolops.GetDistanceMatrix(m) #np.zeros((d,d)); #Adjacency matrix
tempBO = np.zeros((d,d))
tempAR = np.zeros((d,d))
num_features = 11
tempfeatures = [[0]*num_features for _ in range(d)]; # d=#nodes, f=#features available
for atom in vertices:
# Get features of the atom
v_i = atom.GetIdx()
atomic_num = atom.GetAtomicNum()
degree = atom.GetDegree()
valence = atom.GetTotalValence()
hybrid = int(atom.GetHybridization())
atom_aromatic = atom.GetIsAromatic()
ring = atom.IsInRing()
rad = atom.GetNumRadicalElectrons()
pc = atom.GetDoubleProp("_GasteigerCharge")
# Get bonds
linked_atoms =[x.GetIdx() for x in atom.GetNeighbors()]
# Populate adjacency matrix
tempBO[v_i][v_i] = 10 #arbitrarily large number to indicate connectivity to self
for v_j in linked_atoms:
bond_order = m.GetBondBetweenAtoms(v_i,v_j).GetBondTypeAsDouble()
#bond_length = m.GetBondBetweenAtoms(v_i,v_j).GetBondLength()
bond_aromatic = m.GetBondBetweenAtoms(v_i,v_j).GetIsAromatic()
tempBO[v_i][v_j] = bond_order
tempAR[v_i][v_j] = bond_aromatic
# Write features
tempfeatures[v_i][0] = atomic_num
tempfeatures[v_i][1] = int(atomic_num==1) #H
tempfeatures[v_i][2] = int(atomic_num==6) #C
tempfeatures[v_i][3] = int(atomic_num==7) #N
tempfeatures[v_i][4] = int(atomic_num==8) #O
tempfeatures[v_i][5] = int(atomic_num==9) #F
tempfeatures[v_i][6] = int(atom_aromatic)
tempfeatures[v_i][7] = hybrid
tempfeatures[v_i][8] = degree
tempfeatures[v_i][9] = valence
tempfeatures[v_i][10] = atom.GetDoubleProp("_GasteigerCharge")
As[0].append(sp.coo_matrix(tempA))
As[1].append(sp.coo_matrix(tempBO))
As[2].append(sp.coo_matrix(tempAR))
if (num_molecules == 0):
sizes = sizes + [d-1]
else:
sizes = sizes + [d]
num_molecules=num_molecules+1
if (num_molecules % 5000 == 0):
print("On the "+str(num_molecules)+"th molecule")
features+=tempfeatures
return features, target, As, sizes, num_molecules
except Exception as e:
# Write exception to file
print(str(e))
# with open("analysis/problem_files.txt","a+") as file:
# file.write(str(num_molecules)+" : "+str(url)+ " " + str(e) + "\n")
return features, target, A, sizes, num_molecules
def add_sample(url, features, target, As, sizes, num_molecules, elements_info):
#extract information from xyz file
#try:
if (True):
properties = []
with open(url, 'r') as file:
for row in file:
properties += row.split()
SMILES = properties[-4]
INCHI = properties[-2]
m = Chem.MolFromSmiles(SMILES)
m = Chem.AddHs(m)
vertices = m.GetAtoms()
#edges = m.GetBonds()
d = len(vertices)
partial_charges = properties[22:(23 + 5 * (d - 1)):5]
atomic_number_mean = elements_info[1]
atomic_number_stdev = elements_info[2]
#pc = m.ComputeGasteigerCharges()
#print(pc,partial_charges)
# Targets
dipole_moment = float(properties[6])
polarizability = float(properties[7])
h**o = float(properties[8])
lumo = float(properties[9])
gap = float(properties[10])
r2 = float(properties[11])
zpve = float(properties[12])
U0 = float(properties[13])
internal_energy = float(properties[14])
enthalpy = float(properties[15])
free_nrg = float(properties[16])
heat_capacity = float(properties[17])
target.append([heat_capacity])
#target.append([dipole_moment,polarizability,h**o,lumo,gap,
# r2,zpve,U0,internal_energy,enthalpy,free_nrg,heat_capacity])
#populate the adjacency matrix and write features
tempA = rdmolops.GetDistanceMatrix(
m) #np.zeros((d,d)); #Adjacency matrix
tempBO = np.zeros((d, d))
tempAR = np.zeros((d, d))
f = 8
tempfeatures = [[0] * f for _ in range(d)]
# d=#nodes, f=#features available
dic = {'C': 6, 'H': 1, 'O': 8, 'N': 7, 'F': 9}
for atom in vertices:
# Get features of the atom
v_i = atom.GetIdx()
atomic_num = atom.GetAtomicNum()
degree = atom.GetDegree()
valence = atom.GetTotalValence()
hybrid = int(atom.GetHybridization())
atom_aromatic = atom.GetIsAromatic()
ring = atom.IsInRing()
rad = atom.GetNumRadicalElectrons()
# Get bonds
linked_atoms = [x.GetIdx() for x in atom.GetNeighbors()]
# Populate adjacency matrix
tempBO[v_i][
v_i] = 10 #arbitrarily large number to indicate connectivity to self
for v_j in linked_atoms:
bond_order = m.GetBondBetweenAtoms(v_i,
v_j).GetBondTypeAsDouble()
#bond_length = m.GetBondBetweenAtoms(v_i,v_j).GetBondLength()
bond_aromatic = m.GetBondBetweenAtoms(v_i, v_j).GetIsAromatic()
tempBO[v_i][v_j] = bond_order
tempAR[v_i][v_j] = bond_aromatic
# Write features
tempfeatures[v_i][0] = (atomic_num -
atomic_number_mean) / atomic_number_stdev
tempfeatures[v_i][1] = int(atomic_num == 1) #H
tempfeatures[v_i][2] = int(atomic_num == 6) #C
tempfeatures[v_i][3] = int(atomic_num == 7) #N
tempfeatures[v_i][4] = int(atomic_num == 8) #O
tempfeatures[v_i][5] = int(atomic_num == 9) #F
tempfeatures[v_i][6] = int(atom_aromatic)
tempfeatures[v_i][7] = hybrid
#tempfeatures[v_i][8] = ring #float(partial_charges[atom]) #Mulliken partial charge
#tempfeatures[v_i][9] = degree
#tempfeatures[v_i][10] = valence
As[0].append(sp.coo_matrix(tempA))
As[1].append(sp.coo_matrix(tempBO))
As[2].append(sp.coo_matrix(tempAR))
if (num_molecules == 0):
sizes = sizes + [d - 1]
else:
sizes = sizes + [d]
num_molecules = num_molecules + 1
if (num_molecules % 5000 == 0):
print("On the " + str(num_molecules) + "th molecule")
features += tempfeatures
return features, target, As, sizes, num_molecules
def distance_matrix(mol):
""" The topological distance matrix. """
return rdmolops.GetDistanceMatrix(mol)