Molecular Representation Methods and Scaffold Hopping: A Review

Abstract

This review provides a comprehensive overview of recent advances in molecular representation methods and their applications in scaffold hopping. Molecular representation is a fundamental component of computational drug design, as the way molecules are encoded directly impacts the performance of downstream tasks such as virtual screening, molecular property prediction, and molecular generation.

We systematically categorize and discuss various molecular representation approaches, from traditional fingerprint-based methods to modern deep learning-based representations, and examine how these representations can be leveraged for scaffold hopping — the discovery of structurally novel compounds with similar biological activity.

Key Topics Covered

Molecular Representation Methods

• Traditional Methods: Molecular fingerprints (ECFP, MACCS), molecular descriptors
• Graph-Based Methods: Graph neural networks for molecular graphs
• 3D-Based Methods: Conformational and spatial molecular representations
• Sequence-Based Methods: SMILES-based language models
• Multimodal Methods: Integration of multiple data modalities (structure, phenotype, text)

Scaffold Hopping Applications

• Ligand-Based Scaffold Hopping: Using molecular similarity in learned representation spaces
• Structure-Based Approaches: Combining molecular representations with target structure information
• Generative Approaches: Molecular generation with scaffold constraints
• Case Studies: Successful examples of AI-driven scaffold hopping in drug discovery

BibTeX

@article{wang2025review,
  title={Recent advances in molecular representation methods and their applications in scaffold hopping},
  author={Wang, Shihang and Zhang, Ran and Li, Xiangcheng and Bai, Fang},
  journal={npj Drug Discovery},
  volume={2},
  number={14},
  year={2025},
  doi={10.1038/s44386-025-00017-2}
}