MULocDeep: A deep-learning framework for protein subcellular and suborganellar localization prediction with...

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Protein localization is crucial for understanding protein function and biological mechanisms.
• Accurate prediction of protein localization aids in deciphering cellular processes.
• Existing methods often struggle with predicting multiple localizations or fine-grained suborganellar locations.

Purpose of the Study:

• To develop a general deep learning framework, MULocDeep, for predicting multiple protein localizations.
• To predict protein localization at both subcellular and suborganellar levels.
• To provide insights into protein sorting mechanisms by assessing amino acid contributions.

Main Methods:

• Developed MULocDeep, a deep learning-based localization prediction framework.
• Collected and curated a comprehensive dataset with 44 suborganellar localization annotations across 10 major subcellular compartments.
• Experimentally generated and publicly released an independent dataset of mitochondrial proteins from *Arabidopsis thaliana*, *Solanum tuberosum*, and *Vicia faba*.

Main Results:

• MULocDeep demonstrates superior performance compared to other major methods for both subcellular and suborganellar localization prediction.
• The framework successfully predicts multiple localization sites for proteins.
• MULocDeep identifies specific amino acid contributions, offering mechanistic insights into protein sorting.

Conclusions:

• MULocDeep represents a significant advancement in predicting protein localization, particularly at the suborganellar level.
• The tool provides valuable insights into protein sorting mechanisms and localization motifs.
• The publicly available datasets and web server (http://mu-loc.org) facilitate further research in protein localization.