Analyzing Learned Molecular Representations for Property Prediction

TLDR

Neural networks using fixed fingerprints or graph convolutional networks have shown promise for molecular property prediction, yet it remains unclear which approach generalizes better to new chemical space and how they perform in industrial settings. This study benchmarks fingerprint‑based and graph‑convolutional models across 19 public and 16 proprietary industrial datasets to evaluate their generalization and industrial applicability. The authors develop a new graph convolutional model that consistently matches or surpasses fixed‑descriptor and prior GCN baselines on both public and proprietary datasets. Empirical results show the new model delivers significant performance gains over current industrial workflows, though reproducibility of representation‑based approaches remains limited.

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.