from sklearn.decomposition import PCA
import rdkit.Chem as Chem
from rdkit.Chem import Draw
if __name__ == '__main__':
root_dir = '../../../big_data'
# result_dir = os.path.join(root_dir, 'figures', 'chapter3_figures')
# sub_dir = '06_model_Parallel2vec' # 05_model_Tandem2vec or 06_model_Parallel2vec
# input_dir = os.path.join(root_dir, sub_dir)
need_plot_md = ['nN', 'nS', 'nBondsD', 'naRing']
frag2info = pd.read_csv(os.path.join(root_dir, '03_fragment',
'frag_smiles2md.csv'),
index_col=0)
frag2info = frag2info.loc[:, need_plot_md]
print_df(frag2info)
# if not os.path.exists(result_dir):
# os.makedirs(result_dir)
for frag_sentence_type in ['parallel']: # only plot parallel
print(
'Deal with fragment sentence type: {}'.format(frag_sentence_type))
# frag_sentence_type = 'tandem' # parallel or tandem
if frag_sentence_type == 'parallel':
sub_dir2 = '06_model_Parallel2vec'
else:
sub_dir2 = '05_model_Tandem2vec'
# sub_dir_tandem = '05_model_Tandem2vec'
# sub_dir_parallel = '06_model_Parallel2vec'
# minn = 1
# maxn = 2
# find aromatic ring and corresponding non-aromatic ring pair
bond_pair = find_aromatic_non_aroma_ring_pair(frag_df=frag_info)
with open(bond_pair_file_path, 'a') as f_handle:
if bond_pair:
for bp in bond_pair:
# print(bp)
f_handle.write('\t'.join(list(bp) + ['aromatic_ring']) +
'\n')
frag_pairs = pd.read_csv(bond_pair_file_path, sep='\t')
frag2vec = pd.read_csv(frag_smiles2vec_file_path, index_col='fragment')
pca = PCA(n_components=2)
x_reduced_pca = pd.DataFrame(data=pca.fit_transform(frag2vec),
index=frag2vec.index)
print('>>> x_reduced_pca')
print_df(x_reduced_pca)
# frag_pairs = frag_pairs.loc[frag_pairs['keep'] == 1]
print('>>> frag_pairs')
print_df(frag_pairs)
for bond_type in frag_pairs['bond_type'].unique():
print('>>> Deal with {}...'.format(bond_type))
plt.figure(figsize=(8, 6))
current_bond_pairs = frag_pairs.loc[frag_pairs['bond_type'] ==
bond_type].copy()
if current_bond_pairs.shape[0] > 100:
current_bond_pairs = current_bond_pairs.sample(n=100,
random_state=42)
current_frag1 = current_bond_pairs.loc[:, 'frag1']
current_frag2 = current_bond_pairs.loc[:, 'frag2']
print(frag_vec_file_path)
get_mol_vec(frag2vec_file_path=frag_vec_file_path,
data_set=cid2frag_smiles_file_path,
result_path=mol_vector_file_path)
train_set_file_path = os.path.join(result_dir, 'train_set.csv')
test_set_file_path = os.path.join(result_dir, 'test_set.csv')
if not (os.path.exists(train_set_file_path)
and os.path.exists(test_set_file_path)):
print('Split down-sampled dataset...')
split_data = split_data_set(
os.path.join(root_dir, subdir1, down_sampled_mol_file))
train_set = split_data['train_set']
test_set = split_data['test_set']
print('>>> Training set')
print_df(train_set)
print('>>> Test set')
print_df(test_set)
train_set.to_csv(train_set_file_path)
test_set.to_csv(test_set_file_path)
# --------------------------------------------------------------------------------------
# train model
print('Start to train MLP model...')
train_set = pd.read_csv(train_set_file_path, index_col='cid')
selected_mol2md = pd.read_csv(selected_mol2md_file_path, index_col='cid')
md = get_ordered_md()
selected_mol2md = selected_mol2md.loc[:, md].copy()
y = selected_mol2md.loc[selected_mol2md.index.isin(train_set.index)]
for frag_type in ['tandem', 'parallel',
'random']: # ['tandem', 'parallel']
def archive():
# deal with y
print(' > Start to deal with y...')
frag2md_info_file = ''
frag2vec_file = ''
frag2md_info = pd.read_csv(os.path.join(root_dir, frag2md_info_file),
index_col='fragment')
if model_type == 'classification':
frag2md_info[frag2md_info >= 1] = 1
all_x = pd.read_csv(os.path.join(root_dir, frag2vec_file), index_col=0)
x = all_x.loc[all_x.index != 'UNK'].copy()
y = frag2md_info.loc[x.index, SELECTED_MD].copy()
# train model
print(' > Start to train model...')
for md in SELECTED_MD:
if md == 'naRing':
result_dir_new = os.path.join(result_dir, md + '_aromaticity')
m_part1 = nn_model_regression(x=x,
y=y.loc[:, [md]],
epochs=100,
result_dir=result_dir_new,
callback=True)
else:
result_dir_new = os.path.join(result_dir, md)
m_part1 = nn_model_regression(x=x,
y=y.loc[:, [md]],
epochs=100,
result_dir=result_dir_new,
callback=True)
# get new frag_id2vec in 30D
frag2vec_30d = pd.DataFrame(data=m_part1.predict(all_x),
index=all_x.index)
print_df(frag2vec_30d)
frag2vec_new_fp = os.path.join(result_dir_new, 'frag2vec_new_30d.csv')
frag2vec_30d.to_csv(frag2vec_new_fp)
# plot by t-SNE
x_reduced = reduce_by_tsne(frag2vec_30d)
x_reduced = pd.DataFrame(data=x_reduced, index=frag2vec_30d.index)
print(' >Start to plot t-SNE vis of fragment vector...')
# save_fig_path = os.path.join('./chapter4_figure/', 't-SNE_vis_new_30d.png')
need_plot_md = [md] + list(
np.random.choice(SELECTED_MD, 3, replace=False))
fig = show_each_md(x_reduced=x_reduced,
frag_info=frag2md_info.loc[:, need_plot_md])
fig.savefig(os.path.join(result_dir_new,
't-SNE_vis_sorted_by_{}.png'.format(md)),
dpi=200)
# plot top n fragment
topn = 4
print(
' >Start to plot top {} nearest neighbor of selected fragment vector...'
.format(topn))
q_frags = [
"C1=COCO1", "C1=CCNN=C1", "C1=CCC1", "OBr", "S=S", "C1#CNCC1"
]
q_mol2vec = frag2vec_30d.loc[q_frags, :]
nn = find_nearest_neighbor(training_mol_vec_fp=frag2vec_new_fp,
query_mol_vec_df=q_mol2vec,
top_n=topn,
mol2md_fp='')
# plot
smiles_list = []
dis = []
legends = []
for inx in range(len(q_frags)):
smiles_list += [
i.split(": ")[0] for i in nn[inx][q_frags[inx]].split('; ')
]
dis += [
str('{:.8f}').format(float(i.split(": ")[1]))
for i in nn[inx][q_frags[inx]].split('; ')
]
# print(dis)
# print(inx, smiles_list)
legends += [
'{}({})'.format(smiles_list[i], dis[i])
for i in range(len(smiles_list))
]
fig = draw_multiple_mol(smiles_list=smiles_list,
mols_per_row=topn,
legends=legends)
# print(type(fig))
# print(fig)
save_fig(fig,
file_path=os.path.join(
result_dir_new,
'top{}_nearest_neighbor_sorted_by_{}.svg'.format(
topn, md)))
cid_list_new = {}
with open(cid2smiles_file_name, 'r') as f:
for i in tqdm(f):
cid = i.split('\t')[0]
if cid != 'cid':
cid_list_new[int(cid)] = 1
cid2md.loc[list(cid_list_new.keys()), :].to_csv(cid2md_file_path_new,
index_label='cid')
print('Start to class molecules by combination by MD...')
class_by_md_combination_file_path = os.path.join(
current_dir, 'class_by_md_combination.csv')
if not os.path.exists(class_by_md_combination_file_path):
cid2md = pd.read_csv(cid2md_file_path_new, index_col='cid')
# assert cid2md.shape[1] == 9
print_df(cid2md)
class_by_md_comb = get_class_md_combination(cid2md)
print(' >check again')
print(sum(class_by_md_comb.index.isin(cid2md.index)))
print_df(class_by_md_comb)
class_by_md_comb.to_csv(class_by_md_combination_file_path,
index_label='cid')
else:
print('>>> Using previous result...')
# class_by_md_comb = pd.read_csv(class_by_md_combination_file_path, index_col=0)
print('Start to down-sampling...')
selected_cid2md_class_file_path = os.path.join(
current_dir, 'selected_cid2md_class.csv')
if not os.path.exists(selected_cid2md_class_file_path):
down_sampling_mol(class_by_md_combination_file_path,
result_dir=current_dir,