PSPI: A deep learning approach for prokaryotic small protein identification

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Small Proteins (SPs) play crucial roles in cellular processes, including immunity and communication.
• Current computational tools for SP identification are limited, especially for prokaryotes, and exhibit suboptimal performance.
• A need exists for efficient and accurate methods to identify prokaryotic SPs.

Purpose of the Study:

• To introduce PSPI, a novel deep learning-based approach for predicting prokaryotic Small Proteins (SPs).
• To evaluate PSPI's performance against existing tools for both prokaryotic and eukaryotic SP identification.
• To highlight the utility of PSPI in advancing the study of SPs.

Main Methods:

• Development of PSPI, a deep learning model specifically designed for prokaryotic SP prediction.
• Incorporation of (n, k)-mers into the PSPI model to enhance feature representation.
• Comparative analysis of PSPI against three existing SP identification tools.

Main Results:

• PSPI demonstrated high accuracy in predicting prokaryotic SPs, including those from the human metagenome.
• PSPI outperformed existing tools in speed, precision, sensitivity, and specificity for both prokaryotic and eukaryotic SPs.
• The inclusion of (n, k)-mers significantly improved PSPI's predictive performance, suggesting the importance of short linear motifs in SPs.

Conclusions:

• PSPI is an effective and efficient tool for identifying prokaryotic Small Proteins (SPs).
• The model's performance suggests that short linear motifs are important features for SP identification.
• PSPI provides a valuable resource for researchers studying SPs and can be adapted for other SP identification tasks.