FACT: Functional Group Alignment and Consistency in Token Space for Structure-aware Molecular Representation Learning

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Science & Research — Life Sciences & Biology, Mathematics & Computational Sciences · Depth: Expert, short

Summary

FACT (Functional Group Alignment and Consistency in Token Space) is an end-to-end framework designed to enhance structure-aware SMILES-based molecular representation learning. It specifically addresses two key challenges: the difficulty in incorporating functional group (FG) information into SMILES models due to a lack of explicit alignment between graph-defined FG atom sets and sequence tokens, which prevents complete substructure masking, and the inconsistency of FG representations arising from multiple valid SMILES forms for a single molecule. FACT tackles these issues by implementing an atom-token alignment module for comprehensive FG span masking during pre-training. Additionally, it enforces FG consistency across various SMILES forms during the fine-tuning phase. Evaluated on MoleculeNet benchmarks, FACT demonstrated state-of-the-art or competitive performance across eight distinct tasks, affirming the efficacy of its alignment and consistency learning approach for molecular representation.

Key takeaway

For AI Scientists or Research Scientists developing molecular representation models, FACT offers a clear path to overcome challenges with functional group integration. If your current SMILES-based models struggle with substructure masking or inconsistent representations from varied SMILES forms, consider implementing atom-token alignment during pre-training and consistency learning during fine-tuning. This approach, demonstrated to achieve state-of-the-art performance on MoleculeNet, can significantly improve the accuracy of molecular property predictions.

Key insights

FACT improves molecular representation by aligning functional groups with SMILES tokens and ensuring consistency across forms.

Principles

• Explicit atom-token alignment is crucial for substructure masking.
• Consistent functional group representation enhances model robustness.
• Multiple SMILES forms necessitate consistency learning.

Method

FACT uses an atom-token alignment module for complete functional group span masking during pre-training. It then enforces functional group consistency across different SMILES forms during fine-tuning to improve representation learning.

In practice

• Apply FG span masking for better substructure learning.
• Use consistency learning for robust SMILES representations.
• Improve molecular property prediction tasks.

Topics

• Molecular Representation Learning
• Functional Groups
• SMILES
• Atom-Token Alignment
• Substructure Masking
• MoleculeNet Benchmarks

Best for: AI Scientist, Research Scientist

Related on AIssential

• Agentic Molecular Recovery via Molecule-Aware Exploration — Molecular recovery should prioritize identity preservation and semantic alignment over mere validity restoration for invalid SMILES.
• SRA: Span Representation Alignment for Large Language Model Distillation — Aligning tokenizer-agnostic spans via their attention-weighted Center of Mass improves cross-tokenizer knowledge distillation.
• Author Correction: Biophysical prediction of protein–peptide interactions and signaling networks using machine learning — Data quality is paramount for accurate machine learning models in biophysical predictions.
• Learning Topological Representations for Molecular Dynamics — Masked Flood complex enhances persistent homology for molecular dynamics by incorporating protein-specific structural bias.
• MōLe-Λ: Learning the Coupled-Cluster Response State for Energies, Gradients, and Properties — MōLe-Λ efficiently predicts full CCSD response states by learning both right- and left-hand amplitudes, expanding accessible quantum chemistry properties.
• Augmenting Molecular Language Models with Local $n$-gram Memory — Integrating local $n$-gram memory via MolGram efficiently resolves the locality gap in molecular language models, improving performance with fewer parameters.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.