Graph convolution regression with deepchem

Somedays ago, I posted blog about deepchem. I am still playing with deepchem. Today I tried to use graph convolution regression model.
Deepchem provided Graph convolution Regressor. Cool.
I used solubility data provided from AstraZeneca. https://www.ebi.ac.uk/chembl/assay/inspect/CHEMBL3301364
My test code is following. Almost same as deepchem”s example code.
CSVLoader method is very useful because it can not only read data but also calculate graph feature of each molecule.
Next, Define of Graph convolution network.

import tensorflow as tf
import deepchem as dc
import numpy as np
graph_featurizer = dc.feat.graph_features.ConvMolFeaturizer()
loader = dc.data.data_loader.CSVLoader( tasks=['LogS'], smiles_field="CANONICAL_SMILES", id_field="CMPD_CHEMBLID", featurizer=graph_featurizer )
dataset = loader.featurize( './bioactivity.csv' )

splitter = dc.splits.splitters.RandomSplitter()
trainset,testset = splitter.train_test_split( dataset )

hp = dc.molnet.preset_hyper_parameters
param = hp.hps[ 'graphconvreg' ]
print(param['batch_size'])
g = tf.Graph()
graph_model = dc.nn.SequentialGraph( 75 )
graph_model.add( dc.nn.GraphConv( int(param['n_filters']), 75, activation='relu' ))
graph_model.add( dc.nn.BatchNormalization( epsilon=1e-5, mode=1 ))
graph_model.add( dc.nn.GraphPool() )
graph_model.add( dc.nn.GraphConv( int(param['n_filters']), int(param['n_filters']), activation='relu' ))
graph_model.add( dc.nn.BatchNormalization( epsilon=1e-5, mode=1 ))
graph_model.add( dc.nn.GraphPool() )
graph_model.add( dc.nn.Dense( int(param['n_fully_connected_nodes']), int(param['n_filters']), activation='relu' ))
graph_model.add( dc.nn.BatchNormalization( epsilon=1e-5, mode=1 ))
#graph_model.add( dc.nn.GraphGather(param['batch_size'], activation='tanh'))
graph_model.add( dc.nn.GraphGather( 10 , activation='tanh'))

with tf.Session() as sess:
    model_graphconv = dc.models.MultitaskGraphRegressor( graph_model,
                                                      1,
                                                      75,
                                                     batch_size=10,
                                                     learning_rate = param['learning_rate'],
                                                     optimizer_type = 'adam',
                                                     beta1=.9,beta2=.999)
    model_graphconv.fit( trainset, nb_epoch=30 )

train_scores = {}
regression_metric = dc.metrics.Metric( dc.metrics.pearson_r2_score, np.mean )
train_scores['graphconvreg'] = model_graphconv.evaluate( trainset,[ regression_metric ]  )
p=model_graphconv.predict( testset )

print(train_scores) 

Next run the code.

root@08d8f729f78b:/deepchem/pen_test# python graphconv_test.py > datalog

And datalog file is….

Loading raw samples now.
shard_size: 8192
About to start loading CSV from ./bioactivity.csv
Loading shard 1 of size 8192.
Featurizing sample 0
Featurizing sample 1000
...
Starting epoch 29
On batch 0
On batch 50
On batch 100
computed_metrics: [0.52744994044080606]
{'graphconvreg': {'mean-pearson_r2_score': 0.52744994044080606}}

r2 score is still row, but I think it can improve by change of nb_epochs.

All sample code was uploaded to github.
https://github.com/iwatobipen/deeplearning/blob/master/datalog