Identifying subcellular localizations of mammalian protein complexes based on graph theory with a random forest...

Area of Science:

• Computational biology
• Bioinformatics
• Systems biology

Background:

• Identifying protein complex subcellular localization is crucial for understanding human health, disease mechanisms, diagnosis, and therapy.
• Experimental methods for determining protein complex localization lag behind the rapid accumulation of data.
• There is a need for computational methods to rapidly and reliably predict subcellular localizations of protein complexes.

Purpose of the Study:

• To develop a novel computational method for predicting the subcellular localizations of mammalian protein complexes.
• To utilize graph theory and a random forest algorithm for this prediction task.
• To provide a high-throughput tool that complements existing experimental techniques.

Main Methods:

• Protein complexes are modeled as weighted graphs, with nodes representing proteins and edges representing protein-protein interactions.
• Topological structure features are extracted from these graphs to characterize protein complexes.
• A random forest algorithm is employed to build a predictive model based on these features.

Main Results:

• The method achieved high prediction accuracies on a training set (84.78% for plasma membrane/membrane attached, 71.30% for cytoplasm, 82.00% for nucleus) via 10-fold cross-validation.
• Independent test set accuracies were also high (81.31% for plasma membrane/membrane attached, 69.95% for cytoplasm, 81.00% for nucleus).
• These results demonstrate state-of-the-art performance for the proposed computational method.

Conclusions:

• The developed method provides a rapid and reliable approach for predicting mammalian protein complex subcellular localization.
• This computational tool can serve as a valuable high-throughput resource, augmenting experimental localization studies.
• The findings contribute to a better understanding of protein functions in human health and disease.