DeepLoc 2.0: multi-label subcellular localization prediction using protein language models

Area of Science:

• Proteomics
• Bioinformatics
• Computational Biology

Background:

• Protein subcellular localization is crucial for understanding protein function and cellular processes.
• Accurate prediction of protein localization aids in proteomics research and drug discovery.
• Existing tools often lack multi-localization prediction capabilities and interpretability.

Purpose of the Study:

• To introduce DeepLoc 2.0, an improved tool for predicting protein subcellular localization.
• To enhance prediction performance and interpretability compared to previous versions.
• To enable multi-localization prediction for eukaryotic and human proteins.

Main Methods:

• Curated eukaryotic and human multi-location protein datasets with homology partitioning.
• Utilized a pre-trained protein language model for sequence-based predictions.
• Incorporated attention mechanisms for interpretability and prediction of sorting signals.

Main Results:

• Achieved state-of-the-art performance in DeepLoc 2.0.
• Demonstrated improved accuracy using sequence input over protein profiles.
• Provided attention outputs correlating with sorting signal positions.
• Successfully predicted nine types of protein sorting signals with high accuracy.

Conclusions:

• DeepLoc 2.0 offers a significant advancement in predicting protein subcellular localization.
• The tool provides enhanced performance, interpretability, and multi-localization capabilities.
• The integration of protein language models and attention mechanisms represents a novel approach in the field.