Combining heterogeneous data sources for accurate functional annotation of proteins

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate protein function prediction is crucial for biological research.
• Integrating heterogeneous data sources is challenging due to species-specific data limitations.
• Existing methods often rely solely on sequence-based features, limiting comprehensive analysis.

Purpose of the Study:

• To present a multi-view extension of the GOstruct framework for enhanced protein function annotation.
• To enable learning from disparate, heterogeneous data sources using a kernel-based approach.
• To improve the accuracy and utility of protein function prediction by leveraging diverse biological data.

Main Methods:

• Developed a multi-view extension to the GOstruct structured-output framework.
• Integrated diverse data sources by representing each as a kernel within the framework.
• Evaluated the framework's performance against sequence-based and within-species models.

Main Results:

• The multi-view GOstruct framework successfully utilized all available heterogeneous information.
• Achieved superior performance compared to sequence-based models trained across species.
• Demonstrated better results than models trained on collections of data within a single species.

Conclusions:

• The multi-view GOstruct framework offers a robust approach for integrating heterogeneous data for protein function prediction.
• The enhanced natural language processing component achieves performance comparable to sequence information.
• The framework provides a significant advancement in computational approaches to functional genomics.