3D-Mol: A Novel Contrastive Learning Framework for Molecular Property Prediction with 3D Information

What is it about?

Molecular property prediction, crucial for early drug candidate screening and optimization, has seen advancements with deep learning-based methods. While deep learning-based methods have advanced considerably, they often fall short in fully leveraging 3D spatial information. Specifically, current molecular encoding techniques tend to inadequately extract spatial information, leading to ambiguous representations where a single one might represent multiple distinct molecules. Moreover, existing molecular modeling methods focus predominantly on the most stable 3D conformations, neglecting other viable conformations present in reality. To address these issues, we propose 3D-Mol, a novel approach designed for more accurate spatial structure representation. It deconstructs molecules into three hierarchical graphs to better extract geometric information. Additionally, 3D-Mol leverages contrastive learning for pretraining on 20 million unlabeled data, treating their conformations with identical topological structures as weighted positive pairs and contrasting ones as negatives, based on the similarity of their 3D conformation descriptors and fingerprints. We compare 3D-Mol with various state-of-the-art baselines on 7 benchmarks and demonstrate our outstanding performance.

Featured Image

Photo by Diana Polekhina on Unsplash

Photo by Diana Polekhina on Unsplash

Why is it important?

Molecular property prediction can effectively accelerate drug discovery by prioritizing promising compounds, streamlining drug development and increasing success rates. Moreover, it contributes to the comprehension of structure-activity relationships by demonstrating the influence of particular features on molecular interactions and other biological effects. Recently, deep learning methods have significantly advanced molecular property prediction, providing enhanced accuracy and deeper insights into complex molecular behaviors. The integration of 3D molecular information, which includes a comprehensive view of molecular structures, significantly enhance the model’s understanding of molecular properties and interactions. However, the expensive and time-consuming experiments result in the scarcity of labeled data, which significantly constrains the capacity of deep learning methods to extract 3D spatial information.

Perspectives