A random forest model for predicting exosomal proteins using evolutionary information and motifs.

Area of Science:

• Biochemistry
• Bioinformatics
• Proteomics

Background:

• Non-invasive diagnostics and therapies are essential for patient well-being.
• Exosomal proteins are key biomarkers for developing advanced diagnostic and therapeutic strategies.
• Predicting exosomal proteins aids in identifying potential targets for non-invasive medical applications.

Purpose of the Study:

• To develop a robust model for predicting exosomal proteins.
• To overcome limitations of traditional similarity-based methods for exosomal protein identification.
• To create a tool for scientists to predict and discover exosomal proteins and their motifs.

Main Methods:

• Utilized a non-redundant dataset of 5662 proteins (2831 exosomal, 2831 non-exosomal).
• Evaluated Basic Local Alignment Search Tool (BLAST) for exosomal protein prediction.
• Developed and compared machine learning (ML) models using protein compositional and evolutionary features.
• Integrated motif-based analysis with ML approaches to create a hybrid prediction model.

Main Results:

• Standard BLAST method showed insufficient accuracy due to low protein similarity.
• Machine learning models achieved an Area Under the Receiver Operating Characteristic (AUROC) of 0.73.
• Identified sequence-based motifs unique to exosomal proteins.
• The hybrid model achieved a maximum AUROC of 0.85 and Matthews Correlation Coefficient (MCC) of 0.56 on an independent dataset.
• The hybrid model demonstrated superior performance compared to existing methods.

Conclusions:

• A hybrid approach combining ML and motif-based features significantly improves exosomal protein prediction accuracy.
• The developed ExoProPred web server and software provide a valuable resource for exosomal protein research.
• This work facilitates the discovery of exosomal proteins for non-invasive diagnostics and therapies.