Machine Learning Framework for Conotoxin Class and Molecular Target Prediction.

Area of Science:

• Biochemistry and Pharmacology
• Computational Biology
• Neuroscience

Background:

• Conotoxins are potent neurotoxic peptides from cone snail venom with diverse structures and specificities for ion channels and receptors.
• Accurately predicting conotoxin binding targets and toxicities is challenging due to complex structure-function relationships and conformational heterogeneity.
• Previous work showed that including post-translational modifications and collisional cross sections improves prediction accuracy over primary sequence alone.

Purpose of the Study:

• To evaluate the impact of additional structural features on conotoxin class and molecular target prediction.
• To specifically improve the prediction of conotoxins targeting nicotinic acetylcholine receptors (nAChRs).
• To apply dataset balancing techniques like SMOTE-Tomek for enhanced model performance.

Main Methods:

• Utilized machine learning classifiers with combined standard sequence features and advanced structural features (post-translational modifications, collisional cross sections).
• Employed Synthetic Minority Oversampling Technique (SMOTE)-Tomek for dataset balancing and class separation.
• Developed predictive models including SMOTE-Tomek PCA PLR for conotoxin class prediction and SMOTE-Tomek PCA SVM for nAChR target prediction.

Main Results:

• The SMOTE-Tomek PCA PLR model achieved 95.95% overall accuracy for alpha, mu, and omega conotoxin class prediction.
• The SMOTE-Tomek PCA SVM model demonstrated 91.3% overall accuracy in predicting conotoxins that bind to nAChRs.
• High sensitivities were achieved for predicting specific conotoxin classes and nAChR binding, indicating robust model performance.

Conclusions:

• Integrating structural features significantly enhances the predictive power for conotoxin classification and molecular target identification.
• The developed models provide accurate predictions for conotoxin classes and their binding to nAChRs, advancing drug discovery and toxin research.
• SMOTE-Tomek balancing combined with structural features offers a powerful approach for analyzing complex peptide toxin datasets.