def tensorize(tree_batch, vocab, assm=True, if_need_origin_word = False):
set_batch_nodeID(tree_batch, vocab)
smiles_batch = [tree.smiles for tree in tree_batch]
(jtenc_holder,mess_dict),origin_word = JTNNEncoder.tensorize(tree_batch)
mpn_holder = MPN.tensorize(smiles_batch)
if assm is False:
if if_need_origin_word:
return tree_batch, jtenc_holder, mpn_holder, origin_word
else:
return tree_batch, jtenc_holder, mpn_holder
cands = []
batch_idx = []
for i,mol_tree in enumerate(tree_batch):
for node in mol_tree.nodes:
#Leaf node's attachment is determined by neighboring node's attachment
if node.is_leaf or len(node.cands) == 1: continue
cands.extend( [(cand, mol_tree.nodes, node) for cand in node.cands] )
batch_idx.extend([i] * len(node.cands))
jtmpn_holder = JTMPN.tensorize(cands, mess_dict)
batch_idx = torch.LongTensor(batch_idx)
return tree_batch, jtenc_holder, mpn_holder, (jtmpn_holder,batch_idx)
def dfs_assemble(self, y_tree_mess, x_mol_vec_pooled, all_nodes, cur_mol, global_amap, fa_amap, cur_node, fa_node):
fa_nid = fa_node.nid if fa_node is not None else -1
prev_nodes = [fa_node] if fa_node is not None else []
children = [nei for nei in cur_node.neighbors if nei.nid != fa_nid]
neighbors = [nei for nei in children if nei.mol.GetNumAtoms() > 1]
neighbors = sorted(neighbors, key=lambda x:x.mol.GetNumAtoms(), reverse=True)
singletons = [nei for nei in children if nei.mol.GetNumAtoms() == 1]
neighbors = singletons + neighbors
cur_amap = [(fa_nid,a2,a1) for nid,a1,a2 in fa_amap if nid == cur_node.nid]
cands = enum_assemble(cur_node, neighbors, prev_nodes, cur_amap)
if len(cands) == 0:
return None
cand_smiles,cand_amap = zip(*cands)
cands = [(smiles, all_nodes, cur_node) for smiles in cand_smiles]
jtmpn_holder = JTMPN.tensorize(cands, y_tree_mess[1])
fatoms,fbonds,agraph,bgraph,scope = jtmpn_holder
cand_vecs = self.jtmpn(fatoms, fbonds, agraph, bgraph, scope, y_tree_mess[0])
scores = torch.mv(cand_vecs, x_mol_vec_pooled)
_,cand_idx = torch.sort(scores, descending=True)
backup_mol = Chem.RWMol(cur_mol)
#for i in xrange(cand_idx.numel()):
for i in xrange( min(cand_idx.numel(), 5) ):
cur_mol = Chem.RWMol(backup_mol)
pred_amap = cand_amap[cand_idx[i].item()]
new_global_amap = copy.deepcopy(global_amap)
for nei_id,ctr_atom,nei_atom in pred_amap:
if nei_id == fa_nid:
continue
new_global_amap[nei_id][nei_atom] = new_global_amap[cur_node.nid][ctr_atom]
cur_mol = attach_mols(cur_mol, children, [], new_global_amap) #father is already attached
new_mol = cur_mol.GetMol()
new_mol = Chem.MolFromSmiles(Chem.MolToSmiles(new_mol))
if new_mol is None: continue
result = True
for nei_node in children:
if nei_node.is_leaf: continue
cur_mol = self.dfs_assemble(y_tree_mess, x_mol_vec_pooled, all_nodes, cur_mol, new_global_amap, pred_amap, nei_node, cur_node)
if cur_mol is None:
result = False
break
if result: return cur_mol
return None
def dfs_assemble(self, y_tree_mess, x_mol_vecs, all_nodes, cur_mol, global_amap, fa_amap, cur_node, fa_node, prob_decode, check_aroma):
fa_nid = fa_node.nid if fa_node is not None else -1
prev_nodes = [fa_node] if fa_node is not None else []
children = [nei for nei in cur_node.neighbors if nei.nid != fa_nid]
neighbors = [nei for nei in children if nei.mol.GetNumAtoms() > 1]
neighbors = sorted(neighbors, key=lambda x:x.mol.GetNumAtoms(), reverse=True)
singletons = [nei for nei in children if nei.mol.GetNumAtoms() == 1]
neighbors = singletons + neighbors
cur_amap = [(fa_nid,a2,a1) for nid,a1,a2 in fa_amap if nid == cur_node.nid]
cands,aroma_score = enum_assemble(cur_node, neighbors, prev_nodes, cur_amap)
if len(cands) == 0 or (sum(aroma_score) < 0 and check_aroma):
return None, cur_mol
cand_smiles,cand_amap = zip(*cands)
aroma_score = torch.Tensor(aroma_score).cuda()
cands = [(smiles, all_nodes, cur_node) for smiles in cand_smiles]
if len(cands) > 1:
jtmpn_holder = JTMPN.tensorize(cands, y_tree_mess[1])
fatoms,fbonds,agraph,bgraph,scope = jtmpn_holder
cand_vecs = self.jtmpn(fatoms, fbonds, agraph, bgraph, scope, y_tree_mess[0])
scores = torch.mv(cand_vecs, x_mol_vecs) + aroma_score
else:
scores = torch.Tensor([1.0])
if prob_decode:
probs = F.softmax(scores.view(1,-1), dim=1).squeeze() + 1e-7 #prevent prob = 0
cand_idx = torch.multinomial(probs, probs.numel())
else:
_,cand_idx = torch.sort(scores, descending=True)
backup_mol = Chem.RWMol(cur_mol)
pre_mol = cur_mol
for i in range(cand_idx.numel()):
cur_mol = Chem.RWMol(backup_mol)
pred_amap = cand_amap[cand_idx[i].item()]
new_global_amap = copy.deepcopy(global_amap)
for nei_id,ctr_atom,nei_atom in pred_amap:
if nei_id == fa_nid:
continue
new_global_amap[nei_id][nei_atom] = new_global_amap[cur_node.nid][ctr_atom]
cur_mol = attach_mols(cur_mol, children, [], new_global_amap) #father is already attached
new_mol = cur_mol.GetMol()
new_mol = Chem.MolFromSmiles(Chem.MolToSmiles(new_mol))
if new_mol is None: continue
has_error = False
for nei_node in children:
if nei_node.is_leaf: continue
tmp_mol, tmp_mol2 = self.dfs_assemble(y_tree_mess, x_mol_vecs, all_nodes, cur_mol, new_global_amap, pred_amap, nei_node, cur_node, prob_decode, check_aroma)
if tmp_mol is None:
has_error = True
if i == 0: pre_mol = tmp_mol2
break
cur_mol = tmp_mol
if not has_error: return cur_mol, cur_mol
return None, pre_mol