Chemception: A Deep Neural Network with Minimal Chemistry Knowledge Matches the Performance of Expert-developed QSAR/QSPR Models

In the last few years, we have seen the transformative impact of deep in many applications, particularly in speech recognition and computer. Inspired by Google's Inception-ResNet deep convolutional neural network (CNN) for image classification, we have developed "Chemception", a deep CNN for prediction of chemical properties, using just the images of 2D drawings of. We develop Chemception without providing any additional explicit knowledge, such as basic concepts like periodicity, or advanced like molecular descriptors and fingerprints. We then show how can serve as a general-purpose neural network architecture for toxicity, activity, and solvation properties when trained on a database of 600 to 40, 000 compounds. When compared to multi-layer (MLP) deep neural networks trained with ECFP fingerprints, slightly outperforms in activity and solvation prediction and underperforms in toxicity prediction. Having matched the performance expert-developed QSAR/QSPR deep learning models, our work demonstrates the of using deep neural networks to assist in computational chemistry, where the feature engineering process is performed primarily by a learning algorithm.