Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hierarchical Structure-Property Alignment for Data-Efficient Molecular Generation and Editing

Ziyu Fan, Zhijian Huang, Yahan Li, Xiaowen Hu, Siyuan Shen, Yunliang Wang, Zeyu Zhong, Shuhong Liu, Shuning Yang, Shangqian Wu, Min Wu, Lei Deng

TL;DR

Experiments demonstrate that HSPAG captures fine-grained structure-property relationships and supports controllable generation under multiple property constraints and two real-world case studies further validate the editing capabilities of HSPAG.

Abstract

Property-constrained molecular generation and editing are crucial in AI-driven drug discovery but remain hindered by two factors: (i) capturing the complex relationships between molecular structures and multiple properties remains challenging, and (ii) the narrow coverage and incomplete annotations of molecular properties weaken the effectiveness of property-based models. To tackle these limitations, we propose HSPAG, a data-efficient framework featuring hierarchical structure-property alignment. By treating SMILES and molecular properties as complementary modalities, the model learns their relationships at atom, substructure, and whole-molecule levels. Moreover, we select representative samples through scaffold clustering and hard samples via an auxiliary variational auto-encoder (VAE), substantially reducing the required pre-training data. In addition, we incorporate a property relevance-aware masking mechanism and diversified perturbation strategies to enhance generation quality under sparse annotations. Experiments demonstrate that HSPAG captures fine-grained structure-property relationships and supports controllable generation under multiple property constraints. Two real-world case studies further validate the editing capabilities of HSPAG.

Hierarchical Structure-Property Alignment for Data-Efficient Molecular Generation and Editing

TL;DR

Experiments demonstrate that HSPAG captures fine-grained structure-property relationships and supports controllable generation under multiple property constraints and two real-world case studies further validate the editing capabilities of HSPAG.

Abstract

Property-constrained molecular generation and editing are crucial in AI-driven drug discovery but remain hindered by two factors: (i) capturing the complex relationships between molecular structures and multiple properties remains challenging, and (ii) the narrow coverage and incomplete annotations of molecular properties weaken the effectiveness of property-based models. To tackle these limitations, we propose HSPAG, a data-efficient framework featuring hierarchical structure-property alignment. By treating SMILES and molecular properties as complementary modalities, the model learns their relationships at atom, substructure, and whole-molecule levels. Moreover, we select representative samples through scaffold clustering and hard samples via an auxiliary variational auto-encoder (VAE), substantially reducing the required pre-training data. In addition, we incorporate a property relevance-aware masking mechanism and diversified perturbation strategies to enhance generation quality under sparse annotations. Experiments demonstrate that HSPAG captures fine-grained structure-property relationships and supports controllable generation under multiple property constraints. Two real-world case studies further validate the editing capabilities of HSPAG.

Paper Structure

This paper contains 27 sections, 7 equations, 14 figures, 10 tables.

Table of Contents

  1. Dataset
  2. Details of the HSPAG Framework
  3. Construction of the Example and Challenging Sets
  4. Learnable Logit Scaling in InfoNCE Loss
  5. Variational Lower Bound
  6. Limitations
  7. Simulated property annotation.
  8. Missing structural constraints in ligand-based generation.
  9. Molecular generation and Property Prediction Pipeline
  10. Training details and model parameter settings
  11. Compared Baselines
  12. HSPAG Aligns Closely with Real-World Molecular Distributions
  13. Molecular generation process
  14. Evaluation Metrics
  15. HSPAG Enables Accurate Property-Controlled Molecular Generation
  16. ...and 12 more sections

Figures (14)

  • Figure 1: The number of scaffolds with different sampling strategy.
  • Figure 2: (a) Correlation of CVAE LM loss and Availability. (b) Correlation of RMSE and CVAE LM loss. (c) Correlation of CVAE LM loss and VAE LM loss.
  • Figure 3: Distribution comparison across five molecular properties (HeavyAtomNum, QED, SA_Score, MolWt, and SMILES_Length) for different dataset splits.
  • Figure 4: Molecular generation pipeline.
  • Figure 5: Molecular property prediction Pipeline.
  • ...and 9 more figures