Quantitative structure-property relationship modelling for predicting retention indices of essential oils based on an...

Area of Science:

• Computational Chemistry
• Cheminformatics
• Machine Learning

Background:

• Metaheuristic algorithms are increasingly used for complex optimization problems.
• Descriptor selection is crucial for accurate quantitative structure-retention relationship (QSPR) modeling.
• The horse herd optimization algorithm (HOA) is a contemporary metaheuristic with demonstrated potential.

Purpose of the Study:

• To enhance the binary horse herd optimization algorithm (BHOA) for descriptor selection in QSPR.
• To evaluate the efficacy of Z-shape transfer functions (ZTF) in improving BHOA performance.
• To predict essential oil retention indices using optimized QSPR models.

Main Methods:

• Implementation of binary horse herd optimization algorithm (BHOA) for descriptor selection.
• Integration and testing of Z-shape transfer functions (ZTF), particularly ZTF1, with BHOA.
• Validation using mean-squared error (MSE), leave-one-out cross-validation (Q^2), and 5-fold cross-validation.

Main Results:

• Z-shape transfer functions significantly improved BHOA performance in QSPR modeling.
• ZTF1 demonstrated the fastest convergence among binary algorithms tested.
• The optimized models required fewer descriptors and iterations to achieve excellent prediction accuracy for essential oil retention indices.

Conclusions:

• The proposed ZTF, especially ZTF1, effectively enhances BHOA for descriptor selection in QSPR.
• The enhanced BHOA with ZTF provides a robust and efficient method for predicting essential oil retention indices.
• This approach offers improved accuracy and computational efficiency in cheminformatics applications.