FragmentNet: Adaptive Graph Fragmentation for Graph-to-Sequence Molecular Representation Learning

2025-02-03

AI Summary

To address insufficient chemical interpretability and suboptimal performance on large-scale molecular property prediction tasks, this paper proposes a graph-to-sequence adaptive fragmentation modeling framework. Methodologically, it integrates a VQVAE-GCN backbone, spatial positional encoding, global molecular descriptors, and a Transformer architecture. Key contributions include: (1) the first learnable, adaptive graph fragmentation tokenizer that guarantees chemical validity and topological connectivity of fragments; (2) a fragment-level masked modeling pretraining strategy; and (3) support for controllable fragment editing and visualizable property trend analysis. Evaluated on the MoleculeNet benchmark, our model achieves state-of-the-art performance among models of comparable size—outperforming same-scale SOTAs with fewer parameters—while matching or exceeding larger models in accuracy. Crucially, it significantly enhances both predictive performance and chemical interpretability.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Computer Vision: Large Vision Models
• Reasoning under Uncertainty: Graphical Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Molecular property prediction uses molecular structure to infer chemical properties. Chemically interpretable representations that capture meaningful intramolecular interactions enhance the usability and effectiveness of these predictions. However, existing methods often rely on atom-based or rule-based fragment tokenization, which can be chemically suboptimal and lack scalability. We introduce FragmentNet, a graph-to-sequence foundation model with an adaptive, learned tokenizer that decomposes molecular graphs into chemically valid fragments while preserving structural connectivity. FragmentNet integrates VQVAE-GCN for hierarchical fragment embeddings, spatial positional encodings for graph serialization, global molecular descriptors, and a transformer. Pre-trained with Masked Fragment Modeling and fine-tuned on MoleculeNet tasks, FragmentNet outperforms models with similarly scaled architectures and datasets while rivaling larger state-of-the-art models requiring significantly more resources. This novel framework enables adaptive decomposition, serialization, and reconstruction of molecular graphs, facilitating fragment-based editing and visualization of property trends in learned embeddings - a powerful tool for molecular design and optimization.

Problem

• Molecular Graph Representation
• Property Prediction
• Performance Enhancement

Innovation

• Adjustable Graph Slicing
• Masked Fragment Modeling
• VQVAE-GCN