Multi-label learning with fuzzy hypergraph regularization for protein subcellular location prediction.

Area of Science:

• Computational Biology
• Bioinformatics
• Molecular Cell Biology

Background:

• Protein subcellular localization is crucial for understanding protein function and cellular processes.
• Existing computational methods face limitations in accurately predicting locations for proteins with single or multiple cellular residences.

Purpose of the Study:

• To propose a novel hierarchical multi-label learning model (FHML) for enhanced protein subcellular location prediction.
• To address the limitations of current methods in handling both single- and multi-location proteins.

Main Methods:

• Implemented a nonnegative data factorization framework for feature and label space decomposition to extract latent concepts.
• Constructed dual fuzzy hypergraphs to capture high-order relationships in both feature and label spaces.
• Utilized dual fuzzy hypergraph Laplacian regularization for propagating location information from labeled to unlabeled proteins.

Main Results:

• The FHML model demonstrated superior performance on six benchmark datasets compared to state-of-the-art methods.
• The approach effectively leverages both feature and label correlations for improved prediction accuracy.
• Validated the model's efficacy in predicting subcellular locations for proteins with single and multiple cellular residences.

Conclusions:

• The proposed FHML model offers a significant advancement in computational protein subcellular location prediction.
• Exploiting feature and label correlations through dual fuzzy hypergraphs enhances prediction accuracy.
• FHML provides a robust framework for both single- and multi-location protein prediction.