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

By Garrett B. Goh, Charles Siegel, Abhinav Vishnu, Nathan Oken Hodas, Nathan Baker. 2017

182 citations recorded by the source corpus.

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

Subfields: Computational Theory and Mathematics, Materials Chemistry

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Theory and Mathematics

In Materials Chemistry

Citations in the corpus

Mol2vec: Unsupervised Machine Learning Approach with Chemical Intuition (2018, positive): “10 In parallel, natural language processing (NLP) techniques were adopted to learn from classical features, like molecular FPs, 11 SMILES strings, 12 and graph representations of compounds.”
Using Rule-Based Labels for Weak Supervised Learning: A ChemNet for Transferable Chemical Property Prediction (2017, methodology): “In this work, we first evaluated the ChemNet approach, on the Chemception CNN-based model [15, 16].”
How Much Chemistry Does a Deep Neural Network Need to Know to Make Accurate Predictions? (2017, methodology): “[9] A number corresponding to the atomic number (i.”
The rise of deep learning in drug discovery. (2018, result): “[58] applied a CNN on images of 2D drawings of molecules and achieved surprisingly comparable results to DNN models trained on ECFP.”
ChemNet: A Transferable and Generalizable Deep Neural Network for Small-Molecule Property Prediction (2017, methodology): “The dataset splitting steps are identical to that reported previously [9].”
Artificial intelligence to deep learning: machine intelligence approach for drug discovery (2021, background): “…characteristics features of drug compounds such as SMILES2Vec (predicts chemical properties) [339], Chemception (DNN-based prediction of chemical properties) [245], DeepSynergy (prediction of anti-cancer drug synergy with DL) [340], and deepAOT (prediction of compound acute oral toxicity) [341].”
MolGenSurvey: A Systematic Survey in Machine Learning Models for Molecule Design (2022, methodology): “Then standard CNNs [66], [67] are applied on it for representation.”
SMILES2Vec: An Interpretable General-Purpose Deep Neural Network for Predicting Chemical Properties (2017, methodology): “Specifically, we used initial designs that had an embedding size of 40, a convolution layer with 16 filters, and both LSTM/RNN layer with [8, 16, 32, 64, 128, 256] units.”
ToxicBlend: Virtual Screening of Toxic Compounds with Ensemble Predictors (2018, background): “Such approaches draw inspiration from convolutional neural networks for images, even directly from drawings of molecules as in [25].”
Automating drug discovery (2017, background): “Now, ‘deep’ methods based on learning directly from molecular graphs and other physically oriented models of complex molecular objects have been proposed that remove some of this input-level abstraction 190–192 .”
Drug discovery with explainable artificial intelligence (2020, unknown): “(5)) in attention-based graph convolutional architectures.”