Discovering the nuclear localization signal universe through a deep learning model with interpretable attention units

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Nuclear localization signals (NLS) are crucial for protein transport into the nucleus.
• Accurate prediction of NLS is essential for understanding cellular functions.
• Existing NLS prediction methods have limitations in performance and interpretability.

Purpose of the Study:

• To introduce NLSExplorer, an interpretable computational tool for NLS prediction.
• To improve the accuracy and efficiency of NLS detection.
• To explore the NLS landscape and nuclear transport features across diverse species.

Main Methods:

• Developed NLSExplorer, an interpretable approach leveraging protein language models.
• Utilized extracted nuclear-specific site information for enhanced NLS detection.
• Applied NLSExplorer to nucleus-localized proteins in the Swiss-Prot database.

Main Results:

• NLSExplorer achieved over 10% improvement in F1 score compared to existing methods.
• The tool successfully identified NLS and other nuclear transport segments.
• Analysis revealed a potential NLS landscape and features across 416 species.

Conclusions:

• NLSExplorer is a powerful and accurate tool for NLS prediction.
• The study provides insights into the NLS landscape and nuclear transport mechanisms.
• NLSExplorer facilitates the detection of characteristic protein domains and motifs.