Convolutional Embedding of Attributed Molecular Graphs for Physical Property Prediction

TLDR

Learning an expressive molecular representation is central to quantitative structure–activity relationships, yet traditional methods rely on group additivity, empirical measurements, or thousands of descriptors. This study applies a convolutional neural network that treats molecules as undirected graphs with attributed nodes and edges to embed them. By constructing atom‑specific feature vectors from simple atom and bond attributes that capture local chemical environments at varying radii, and by operating directly on the full molecular graph, the model can discover relevant features, with extensions and limitations discussed. The atom featurization preserves molecule‑level spatial information, significantly improving performance, and the models learn to identify atom‑cluster features that predict aqueous solubility, octanol solubility, melting point, and toxicity.

Abstract

The task of learning an expressive molecular representation is central to developing quantitative structure-activity and property relationships. Traditional approaches rely on group additivity rules, empirical measurements or parameters, or generation of thousands of descriptors. In this paper, we employ a convolutional neural network for this embedding task by treating molecules as undirected graphs with attributed nodes and edges. Simple atom and bond attributes are used to construct atom-specific feature vectors that take into account the local chemical environment using different neighborhood radii. By working directly with the full molecular graph, there is a greater opportunity for models to identify important features relevant to a prediction task. Unlike other graph-based approaches, our atom featurization preserves molecule-level spatial information that significantly enhances model performance. Our models learn to identify important features of atom clusters for the prediction of aqueous solubility, octanol solubility, melting point, and toxicity. Extensions and limitations of this strategy are discussed.