exopropred

exopropred predicts exosomal proteins using a hybrid machine-learning and motif-search approach to distinguish exosomal from non-exosomal proteins for biomarker discovery and non-invasive diagnostics and therapeutic research.

Key Features:

• Hybrid model: Combines machine learning and motif-search approaches to enhance prediction accuracy.
• Features used: Employs compositional and evolutionary protein features for model training.
• Training dataset: Trained, tested, and evaluated on 5,662 proteins comprising 2,831 exosomal and 2,831 non-exosomal proteins with ≤40% sequence similarity.
• Performance (ML-only): Machine-learning models achieved an AUROC of 0.73 on internal evaluations.
• Performance (hybrid): The hybrid motif-based and ML-based approach achieved an AUROC of 0.85 and a Matthews correlation coefficient (MCC) of 0.56 on independent datasets.
• Motif discovery: Identifies sequence-based motifs characteristic of exosomal proteins for use in prediction.
• Comparison to BLAST: Addresses cases where Basic Local Alignment Search Tool (BLAST) fails due to low sequence similarity.

Scientific Applications:

• Exosomal protein prediction: Identification of exosomal proteins as candidate biomarkers for non-invasive diagnostics.
• Motif discovery: Discovery of functional sequence motifs within exosomal proteins.
• Biomedical research: Support for research into therapeutic development involving exosome-associated proteins.

Methodology:

Uses compositional and evolutionary protein features to train machine-learning models, applies motif-search to identify sequence-based motifs, and integrates motif-based and ML-based methods in a hybrid approach; trained, tested, and evaluated on a 5,662-protein dataset (2,831 exosomal, 2,831 non-exosomal, ≤40% similarity) with performance measured by AUROC and MCC.