Molecular descriptor subset selection in theoretical peptide quantitative structure-retention relationship model...

Area of Science:

• Computational Chemistry
• Cheminformatics
• Proteomics

Background:

• Quantitative Structure-Retention Relationship (QSRR) models are crucial for predicting peptide retention behavior.
• Molecular descriptor selection is a key step in developing accurate QSRR models.
• Nature-inspired optimization algorithms offer potential for efficient descriptor selection.

Purpose of the Study:

• To compare the performance of five nature-inspired optimization algorithms for molecular descriptor selection in QSRR modeling.
• To develop robust QSRR models for 83 peptides from eight model proteins.
• To evaluate models based on prediction accuracy, computational cost, and the number of selected descriptors.

Main Methods:

• Utilized Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Artificial Bee Colony (ABC), Firefly Algorithm (FA), and Flower Pollination Algorithm (FPA) for descriptor selection.
• Developed QSRR models using Partial Least Squares (PLS) with Root Mean Square Error of Prediction (RMSEP) as the fitness function.
• Compared algorithm performance against interval PLS (iPLS), sparse PLS (sPLS), and full PLS models.

Main Results:

• All five nature-inspired algorithms outperformed iPLS, sPLS, and full PLS models.
• Genetic Algorithm (GA) demonstrated superior performance with the lowest computational cost and highest accuracy (RMSEP 5.534%) using only nine descriptors.
• The GA-QSRR model showed robustness through Y-randomization and external validation (RMSEP 22.030%) on Bacillus subtilis proteome peptides.

Conclusions:

• Genetic Algorithm (GA) is highly effective for molecular descriptor selection in QSRR modeling of peptides.
• The developed GA-QSRR model is accurate, computationally efficient, and robust for proteomics applications.
• The methodology allows for identification of error sources and further application in proteomics research.