Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Molecular Odor Prediction Based on Multi-Feature Graph Attention Networks

HongXin Xie, JianDe Sun, Yi Shao, Shuai Li, Sujuan Hou, YuLong Sun, Jian Wang

TL;DR

This work tackles the Quantitative Structure-Odor Relationship problem by developing a Graph Attention Network-based framework that learns multi-level molecular representations to predict odor descriptors. It integrates local atomic/bond features, SMARTS-based functional groups, and global fingerprints through a Hierarchical Attention Graph Convolutional Network and an Attention Aggregation module, complemented by Adaptive Focal Loss to address label imbalance. On a dataset of 5,788 molecules and 154 descriptors, the proposed method achieves state-of-the-art performance (best AUROC $0.9294$ and F1 $0.4632$), outperforming descriptor-based baselines. The approach demonstrates the practical potential of graph-based, multi-scale representations for QSOR and odor prediction in cheminformatics.

Abstract

Olfactory perception plays a critical role in both human and organismal interactions, yet understanding of its underlying mechanisms and influencing factors remain insufficient. Molecular structures influence odor perception through intricate biochemical interactions, and accurately quantifying structure-odor relationships presents significant challenges. The Quantitative Structure-Odor Relationship (QSOR) task, which involves predicting the associations between molecular structures and their corresponding odors, seeks to address these challenges. To this end, we propose a method for QSOR, utilizing Graph Attention Networks to model molecular structures and capture both local and global features. Unlike conventional QSOR approaches reliant on predefined descriptors, our method leverages diverse molecular feature extraction techniques to automatically learn comprehensive representations. This integration enhances the model's capacity to handle complex molecular information, improves prediction accuracy. Our approach demonstrates clear advantages in QSOR prediction tasks, offering valuable insights into the application of deep learning in cheminformatics.

Molecular Odor Prediction Based on Multi-Feature Graph Attention Networks

TL;DR

This work tackles the Quantitative Structure-Odor Relationship problem by developing a Graph Attention Network-based framework that learns multi-level molecular representations to predict odor descriptors. It integrates local atomic/bond features, SMARTS-based functional groups, and global fingerprints through a Hierarchical Attention Graph Convolutional Network and an Attention Aggregation module, complemented by Adaptive Focal Loss to address label imbalance. On a dataset of 5,788 molecules and 154 descriptors, the proposed method achieves state-of-the-art performance (best AUROC and F1 ), outperforming descriptor-based baselines. The approach demonstrates the practical potential of graph-based, multi-scale representations for QSOR and odor prediction in cheminformatics.

Abstract

Olfactory perception plays a critical role in both human and organismal interactions, yet understanding of its underlying mechanisms and influencing factors remain insufficient. Molecular structures influence odor perception through intricate biochemical interactions, and accurately quantifying structure-odor relationships presents significant challenges. The Quantitative Structure-Odor Relationship (QSOR) task, which involves predicting the associations between molecular structures and their corresponding odors, seeks to address these challenges. To this end, we propose a method for QSOR, utilizing Graph Attention Networks to model molecular structures and capture both local and global features. Unlike conventional QSOR approaches reliant on predefined descriptors, our method leverages diverse molecular feature extraction techniques to automatically learn comprehensive representations. This integration enhances the model's capacity to handle complex molecular information, improves prediction accuracy. Our approach demonstrates clear advantages in QSOR prediction tasks, offering valuable insights into the application of deep learning in cheminformatics.

Paper Structure

This paper contains 10 sections, 16 equations, 3 figures, 2 tables.

Table of Contents

  1. INTRODUCTION
  2. PROPOSED METHOD
  3. Multi-level Feature Extraction
  4. Graph Attentional Aggregation
  5. Loss Function
  6. EXPERIMENTS
  7. Dataset
  8. Method Verification
  9. Ablation Study
  10. CONCLUSION

Figures (3)

  • Figure 1: Overview of our proposed architecture.Our network architecture demonstrates a multi-layer feature extraction and graph attention aggregation framework for efficiently encoding and learning representations of complex input data.
  • Figure 2: The figure illustrates the SMILES representation of molecular structures in the dataset along with their corresponding odor descriptors, highlighting the relationship between chemical structures and olfactory properties.
  • Figure 3: Density distribution of molecular labels in datasets.Eighty percent of the molecules are associated with 1 to 6 odor descriptors, with only a small fraction having more than 10 descriptors.