Are Learned Molecular Representations Ready for Prime Time?

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 15 proprietary industrial datasets spanning a wide variety of chemical endpoints. In addition, we introduce a graph convolutional model that consistently 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.

Building attention and edge message passing neural networks for bioactivity and physical–chemical property prediction

“…In the resulting message passing step, the update of a hidden state has a corresponding direction. This model shares underlying principles with the D-MPNN architecture proposed by Yang et al [35] which also uses directed edges to improve MPNN performance. Their proposed model also injects additional chemical descriptor information alongside the FFNN after the message passing stage.…”

“…We introduce a selection of augmentations to known MPNN architectures, which we refer to as Attention MPNN (AMPNN) and Edge Memory Neural Network (EMNN) [34], and evaluate them against published benchmark results with a range of metrics. The EMNN network shares architectural similarities to the D-MPNN model published by Yang et al [35] that was developed concurrently to this work [36], but the D-MPNN includes additional chemical descriptor information. We applied these two types of neural network to eight datasets from the MoleculeNet [30] benchmark and analyse the performances and offer chemical justification for these results with respect to both architecture and parameter selection.…”

Evaluation of network architecture and data augmentation methods for deep learning in chemogenomics

“…Indeed, we observed that promising modifications of the graph encoder (inspired from other fields of graph learning) to learn molecule descriptors, and of the protein sequence encoder and pairwise representation encoder, as parts of the chemogenomic neuron network, did not yield to significant improvements in prediction performance, on our DrugBank-based dataset. Nevertheless, GNN is a very dynamic field and new developments might lead to improvements in chemogenomics in the future [100].…”

Artificial intelligence in chemistry and drug design

“…In their groundbreaking work, they used focused datasets as small as a series of a dozen chemical derivatives to fit equations that would anticipate fairly complex phenotypic effects such as toxicity [11]. Spurred by this success, a large research area has emerged that focuses specifically on (a) identifying approaches to describe chemical structures in more detail, to capture the characteristics that govern their properties such as pharmacophores and three dimensional structure but also autonomously learned representations [12,13], and (b) derive increasingly complex mathematical relationships that aim at describing the causal relationship between these chemical characteristics and the biological properties of interest for predictive purposes [14,15]. Through an increasing amount of structural information [16], as well as data generation through combinatorial libraries and high-throughput screening, first applications of more complex machine learning models became feasible.…”