Analyzing Learned Molecular Representations for Property Prediction

TL;DR

This paper benchmarks molecular property prediction models, compares neural network approaches, and introduces a new graph convolutional model that improves industrial prediction accuracy.

Contribution

It introduces a novel graph convolutional model that outperforms existing descriptor-based and neural network models across diverse datasets.

Findings

The new model matches or exceeds previous models on public and industrial datasets.

Benchmarking on 35 datasets reveals strengths and limitations of current approaches.

Proposed model offers significant improvements in industrial property prediction workflows.

Abstract

Advancements in neural machinery have led to a wide range of algorithmic solutions for molecular property prediction. Two classes of models in particular have yielded promising results: neural networks applied to computed molecular fingerprints or expert-crafted descriptors, and graph convolutional neural networks that construct a learned molecular representation by operating on the graph structure of the molecule. However, recent literature has yet to clearly determine which of these two methods is superior when generalizing to new chemical space. Furthermore, prior research has rarely examined these new models in industry research settings in comparison to existing employed models. In this paper, we benchmark models extensively on 19 public and 16 proprietary industrial datasets spanning a wide variety of chemical endpoints. In addition, we introduce a graph convolutional model that consistently matches or outperforms models using fixed molecular descriptors as well as previous graph neural architectures on both public and proprietary datasets. Our empirical findings indicate that while approaches based on these representations have yet to reach the level of experimental reproducibility, our proposed model nevertheless offers significant improvements over models currently used in industrial workflows.