ProSE-Pero: Peroxisomal Protein Localization Identification Model Based on Self-Supervised Multi-Task Language...

Area of Science:

• Biochemistry and Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Peroxisomes are vital organelles containing oxidative enzymes.
• Misfolded peroxisomal proteins are linked to various diseases.
• Accurate identification and localization of peroxisomal proteins are crucial for understanding cellular functions and disease mechanisms.

Purpose of the Study:

• To develop a highly accurate model for identifying and localizing peroxisomal proteins.
• To leverage deep representation learning for feature extraction from protein sequences.
• To establish a robust computational tool for advancing peroxisomal research.

Main Methods:

• Employed deep representation learning models for peroxisomal protein feature extraction.
• Utilized SVMSMOTE, SHAP, ANOVA, and LightGBM for feature selection and comparison.
• Trained and validated the ProSE-Pero model using tenfold cross-validation on a dataset of 160 peroxisomal proteins.

Main Results:

• The ProSE-Pero model achieved high performance: 93.37% specificity, 82.41% sensitivity, 95.77% accuracy, and 0.9818 AUC.
• Successfully extended the method to identify plant vacuole proteins with 91.90% accuracy, outperforming the iPVP-DRLF model.
• Demonstrated superior performance in peroxisomal protein localization and identification compared to the In-Pero model.

Conclusions:

• The ProSE-Pero model offers a significant advancement in peroxisomal protein identification and localization.
• The study highlights the effectiveness of the ProSE language model for protein sequence feature extraction.
• The model's generalization capability suggests potential applications for identifying proteins in other organelles like mitochondria and Golgi apparatus.