Soluble N-ethyl-maleimide-sensitive factor Attachment Protein Receptors (SNARE) proteins play critical roles in intracellular trafficking and disease mechanisms, making them relevant targets for precision medicine approaches. However, distinguishing SNARE from NONSNARE proteins remains challenging due to shared sequence motifs and structural similarities. In addition, traditional AI models often lack transparency, which limits their utility in sensitive biomedical contexts. We propose PROTEAN (PROTeins classification with Explainable Artificial iNtelligence), a comprehensive methodology that combines machine learning and explainable AI (XAI) techniques for accurate and interpretable protein classification. The pipeline involves three phases: (1) data preprocessing, conducted on a balanced dataset (D128) of 128 SNARE and NONSNARE protein sequences, (2) model training and evaluation of multiple classifiers, such as Support Vector Machines (SVMs) and Neural Networks (NNs), and (3) interpretation of model decisions via SHAP and LIME XAI models to identify the most influential protein descriptors. The Medium Gaussian SVM outperformed all other models, achieving 92.1% accuracy, 94.8% sensitivity, and 89.5% specificity on a balanced test set of 458 proteins. SHAP and LIME provided consistent and biologically meaningful explanations, highlighting features related to amino acid composition and sequence order as critical for classification. PROTEAN demonstrates that integrating interpretable AI models with well-balanced datasets enhances the performance and transparency of protein classification. These insights are vital for the development of trustworthy AI systems in biomedical applications, offering new tools for disease mechanism analysis, biomarker discovery, and personalized therapeutic strategies.
Biasi et al. (Thu,) studied this question.