ProFun-SOM: Protein Function Prediction for Specific Ontology Based on Multiple Sequence Alignment Reconstruction.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Protein function prediction is vital for understanding evolution and viral mutations.
• Gene Ontology (GO) provides a standardized framework for protein function annotation, creating a hierarchical multilabel classification problem.
• Existing methods face performance bottlenecks due to label dependency and data sparsity in GO annotation, particularly the mixed ontology problem.

Purpose of the Study:

• To propose ProFun-SOM, an innovative multilabel classifier to accurately annotate gene ontologies.
• To overcome the performance bottlenecks in protein function prediction caused by label dependency and data sparsity.
• To leverage multiple sequence alignments (MSAs) for enhanced GO annotation.

Main Methods:

• ProFun-SOM utilizes multiple sequence alignments (MSAs) enhanced through a reconstruction process.
• Integrates enhanced MSAs into a deep learning architecture for prediction.
• Predicts annotations across cellular component, molecular function, biological process, and mixed ontologies.

Main Results:

• ProFun-SOM demonstrates superior performance compared to state-of-the-art methods on CAFA3, SwissProt, and NetGO2 datasets.
• Utilizing MSAs effectively addresses label dependency and data sparsity issues.
• The method alleviates the mixed ontology problem in GO annotation.

Conclusions:

• ProFun-SOM offers an effective solution for accurate protein function prediction.
• The approach successfully overcomes key challenges in hierarchical multilabel classification of GO terms.
• This work provides a valuable tool for advancing research in evolutionary biology and virology.