Mlp4green: A Binary Classification Approach Specifically for Green Odor

Jiuliang Yang, Zhiming Qian, Yi He, Minghao Liu, Wannan Li, Weiwei HanInternational Journal of Molecular Sciences 2024 · DOI

TL;DR

This paper introduces "Mlp4green," a novel machine learning approach for accurately classifying molecules as having a "green odor." By leveraging a Multi-Layer Perceptron (MLP) model, they achieve high predictive accuracy and develop a user-friendly web server, paving the way for more intelligent and standardized research into green odor compounds.

What Problem Does This Paper Address?

The distinct "green odor" emitted by fresh leaves, known for its potential antimicrobial and insect-attracting/repelling properties, has lacked a standardized, efficient method for evaluation and prediction. Traditional experimental approaches are time-consuming and laborious. While machine learning has been applied to predict molecular attributes, its specific application to identifying "green odor" molecules was unexplored, presenting a gap in both fundamental understanding and practical application for fields like drug discovery and pest control.

Methods

The researchers employed a multi-faceted approach:

  • Machine Learning for Classification: They trained and compared four machine learning models – MLP, SVM, RF, and MPNN – for binary classification of green vs. non-green odor molecules. The MLP model demonstrated superior performance across various metrics.
  • Clustering Analysis: Using molecular fingerprint similarity (Dice coefficient), they clustered the identified green odor molecules into groups to explore structural relationships.
  • Molecular Docking: To understand the mechanism of action, they performed molecular docking of representative green odor molecules with the Anopheles gambiae odor-binding protein 1, a known insect olfactory receptor.
  • Quantum Chemical Calculations: HOMO and LUMO orbital diagrams were generated to investigate the electronic properties and reactivity of key molecules.
  • Factor Analysis and Difference Analysis: They analyzed molecular descriptors to identify key features differentiating green from non-green odor molecules.
  • Web Server Development: Based on the best-performing MLP model, they created a publicly accessible web server, "Mlp4green," for predicting green odor.

Datasets

The dataset was compiled from several sources: the LRI & Odour Database-Odour Data, Flavornet, The Good Scents Company Information System, and a previous publication by Hatanaka et al. After initial extraction and cleaning (removing unidentifiable molecules and duplicates), the dataset comprised 587 green odor molecules and 5758 non-green odor molecules. To achieve a 1:1 ratio for training, the non-green odor molecules were stratified and up-sampled to 588.

Key Insights and Ideas

  • MLP's Superiority: The MLP model significantly outperformed SVM, RF, and MPNN in predicting green odor molecules, highlighting its effectiveness for this specific classification task.
  • Structural Signatures of Green Odor: Green odor molecules tend to have lower molecular mass, fewer electrons, and distinct Morgan fingerprints compared to non-green odor molecules. This suggests underlying structural features that contribute to the characteristic scent.
  • Mechanism of Action Insights: Molecular docking revealed strong interactions between green odor molecules and insect olfactory receptors, suggesting their role in insect communication and attraction/repulsion. Specific amino acid residues like TRP114 were identified as crucial for binding.
  • Bridging Prediction and Understanding: The combination of machine learning prediction with molecular docking and quantum chemical analysis provides a more holistic understanding of green odor molecules, moving beyond simple classification.

Key Results

  • The MLP model achieved the highest performance across metrics like Accuracy (94.79%), Precision (81.2%), Average Precision, Matthews coefficient, and Area Under Curve.
  • Clustering identified four main groups of green odor molecules, with representative molecules like 4,5-Dimethylthiazole, 6-Methyl-3-hepten-2-one, Hexyl valerate, and 2-Methyl-4-phenyl-2-butanol.
  • Difference analysis confirmed that green odor molecules generally have lower molecular weight, fewer electrons, and higher Morgan fingerprints compared to non-green odor molecules.
  • A functional web server, "Mlp4green," was developed, offering an accessible tool for predicting green odor based on SMILES input.

Limitations and Future Work

  • Interpretability: The authors acknowledge that deep learning models, while accurate, can suffer from low interpretability. They simplified the research by clustering and focusing on representative molecules, rather than deriving the most essential structural features directly.
  • Mechanism Scope: This study focused on the insect attraction/repulsion aspect of green odor. The authors suggest future work could explore other roles, such as antidepressant or antimicrobial mechanisms, and investigate interactions with human or other relevant receptors.
  • Further Molecular Dynamics: Molecular dynamics simulations could provide deeper insights into the dynamic interactions and mechanisms of action.
  • Feature Extraction: Future research could focus on extracting more fundamental structural features of green odor molecules.

Summary updated

Abstract

Fresh green leaves give off a smell known as "green odor." It has antibacterial qualities and can be used to attract or repel insects. However, a common method for evaluating green odor molecules has never existed. Machine learning techniques are widely used in research to forecast molecular attributes for binary classification. In this work, the green odor molecules were first trained and learned using machine learning methods, and then clustering analysis and molecular docking were performed to further explore their molecular characteristics and mechanisms of action. For comparison, four algorithmic models were employed, MLP performed the best in all metrics, including Accuracy, Precision, Average Precision, Matthews coefficient, and Area under curve. We determined by difference analysis that, in comparison to non-green odor molecules, green odor molecules have a lower molecular mass and fewer electrons. Based on the MLP algorithm, we constructed a binary classification prediction website for green odors. The first application of deep learning techniques to the study of green odor molecules can be seen as a signal of a new era in which green odor research has advanced into intelligence and standardization.
BibTeX
@article{yang2024mlp,
  author = {Jiuliang Yang and Zhiming Qian and Yi He and Minghao Liu and Wannan Li and Weiwei Han},
  title = {{Mlp4green: A Binary Classification Approach Specifically for Green Odor}},
  journal = {International Journal of Molecular Sciences},
  year = {2024},
  doi = {10.3390/ijms25063515},
}
Figure