Can Molecular Evolution Mechanism Enhance Molecular Representation?

TL;DR

The paper addresses how molecular representations can benefit from incorporating evolutionary history rather than relying solely on static molecular graphs. It proposes MEvoN, a network that links ancestral molecules (fewer atoms) to descendants (more atoms) using multi-faceted similarity measures to establish plausible evolutionary paths. By integrating path- and label-aware encodings with a MolE encoder, MEvoN-MPP improves molecular property prediction, achieving about a $32.3\%$ average MAE reduction on QM7/QM9 benchmarks across multiple encoders. This evolutionary perspective deepens the understanding of structure-property relationships and offers practical value for drug discovery and materials optimization.

Abstract

Molecular evolution is the process of simulating the natural evolution of molecules in chemical space to explore potential molecular structures and properties. The relationships between similar molecules are often described through transformations such as adding, deleting, and modifying atoms and chemical bonds, reflecting specific evolutionary paths. Existing molecular representation methods mainly focus on mining data, such as atomic-level structures and chemical bonds directly from the molecules, often overlooking their evolutionary history. Consequently, we aim to explore the possibility of enhancing molecular representations by simulating the evolutionary process. We extract and analyze the changes in the evolutionary pathway and explore combining it with existing molecular representations. Therefore, this paper proposes the molecular evolutionary network (MEvoN) for molecular representations. First, we construct the MEvoN using molecules with a small number of atoms and generate evolutionary paths utilizing similarity calculations. Then, by modeling the atomic-level changes, MEvoN reveals their impact on molecular properties. Experimental results show that the MEvoN-based molecular property prediction method significantly improves the performance of traditional end-to-end algorithms on several molecular datasets. The code is available at https://anonymous.4open.science/r/MEvoN-7416/.

Figures (6)

• Figure 1: Molecular evolutionary network (MEvoN) illustrating the evolution pathway, providing a quantitative method for assessing the magnitude and direction of changes in molecular properties.
• Figure 2: Evolutionary paths and molecular property changes for two molecules from the QM9 dataset. (a) and (c) correspond to 'Cc1ccccc1', while (b) and (d) correspond to 'CC1N=COC1=O'. (a) and (b) illustrate the evolutionary paths of two molecules. (c) and (d) display the corresponding variations in molecular properties.
• Figure 3: The MEvoN method's construction process. The steps are: 1) group molecules by atom count; 2) calculate inter-group similarity; and 3) determine evolutionary relationships based on multiple similarity measures.
• Figure 4: Overview of MEvoN-MPP, which employs the MEvoN method to predict various molecular properties. The process includes evolutionary feature extraction and property prediction.
• Figure 5: Importance of different mutation types in molecular evolution for the QM9 dataset with the GAP as the target property. Each point represents a feature (i.e., mutation type) and its corresponding SHAP value. In this case, the color indicates the feature value and the position along the x-axis reflects the impact on the target property.