SNARER: new molecular descriptors for SNARE proteins classification

Area of Science:

• Protein bioinformatics
• Machine learning in biology
• Molecular descriptor development

Background:

• Soluble NSF Attachment Protein REceptor (SNARE) proteins are crucial for various biological processes, including membrane fusion.
• Accurate classification of SNARE proteins is essential for understanding their diverse functions.
• Existing methods rely on sequence-based features, prompting exploration of novel descriptor types.

Purpose of the Study:

• To introduce and evaluate a novel set of molecular descriptors, SNARER, based on protein physicochemical properties.
• To assess the performance enhancement of SNARE protein binary classifiers using SNARER descriptors.
• To compare the efficacy of SNARER descriptors in conjunction with established feature sets (GAAC, CTDT, CKSAAP, 188D) across different datasets.

Main Methods:

• Construction of balanced (D128) and unbalanced (DUNI) SNARE protein datasets.
• Application of machine learning algorithms: Random Forest (RF), k-Nearest Neighbors (kNN), and AdaBoost.
• Integration of SNARER descriptors with existing feature sets and evaluation using oversampling/subsampling techniques on the unbalanced dataset.

Main Results:

• The addition of SNARER descriptors consistently improved precision across all tested machine learning algorithms.
• On the unbalanced DUNI dataset, SNARER integration led to increased accuracy and sensitivity.
• On the balanced D128 dataset, SNARER descriptors significantly boosted accuracy and specificity, with the best performance (92.3% accuracy, 90.1% sensitivity, 95% specificity) achieved using SNARER + CKSAAP with the RF algorithm.

Conclusions:

• Molecular descriptors incorporating physicochemical and structural protein characteristics demonstrably enhance classification performance.
• Dataset balance is a critical factor influencing prediction accuracy, with balanced training yielding superior results.
• The SNARER descriptor class offers a valuable addition to computational strategies for SNARE protein analysis.