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

Garrett B. Goh, Charles Siegel+3 more|arXiv (Cornell University)|Jun 20, 2017

Computational Drug Discovery Methods45 references115 citations

TL;DR

Chemception trains a deep CNN directly on images of 2D molecular drawings with no explicit chemistry features, and matches or slightly surpasses expert-developed QSAR/QSPR models across several properties.

ABSTRACT

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.

Motivation & Objective

• Assess whether a deep convolutional neural network can predict chemical properties using only images of 2D molecular drawings, without explicit chemical knowledge or descriptors.
• Compare Chemception to expert-developed QSAR/QSPR models (MLPs with ECFP fingerprints) on toxicity, activity, and solvation tasks.
• Evaluate performance across a modest dataset range (roughly 600 to 40,000 compounds).

Proposed method

• Develop Chemception, a deep CNN inspired by Inception-ResNet, that operates on 2D molecular drawings as input.
• Train models on a modest dataset without using explicit chemistry concepts or descriptors.
• Benchmark against multilayer perceptron (MLP) networks using ECFP fingerprints.
• Evaluate on toxicity, activity, and solvation prediction tasks to assess performance gaps and parallels.

Experimental results

Research questions

• RQ1 Can a deep CNN trained solely on 2D molecular drawings achieve comparable performance to expert QSAR/QSPR models that rely on engineered descriptors?
• RQ2 How does Chemception perform on toxicity, activity, and solvation tasks relative to descriptor-based MLP models?
• RQ3 Is minimal chemistry knowledge sufficient for accurate property prediction in QSAR/QSPR contexts?

Key findings

• Chemception slightly outperforms MLPs with ECFP fingerprints for activity and solvation predictions.
• Chemception slightly underperforms MLPs with ECFP fingerprints for toxicity predictions.
• Chemception matches the performance of expert-developed QSAR/QSPR deep learning models using minimal chemistry knowledge.
• The approach demonstrates that deep learning can largely perform feature engineering implicitly for chemical property prediction.
• The study uses datasets in the range of roughly 600 to 40,000 compounds.