Protein function prediction through multi-view multi-label latent tensor reconstruction.

Area of Science:

• Bioinformatics
• Computational Biology
• Proteomics

Background:

• High-throughput sequencing accelerates protein discovery, but experimental functional characterization is limited.
• The functions of most newly discovered proteins remain unknown.
• Accurate, fast, and scalable computational methods are crucial for protein function prediction.

Purpose of the Study:

• To develop an advanced computational method for automatic protein function prediction.
• To leverage multi-view protein features for improved functional annotation.
• To address the challenge of predicting functions for proteins with low sequence similarity or rare annotations.

Main Methods:

• Developed GO-LTR, a multi-view, multi-label prediction model.
• Employed high-order tensor approximation of model weights and non-linear activation functions.
• Learned high-order relationships between multiple protein feature views.

Main Results:

• GO-LTR demonstrates competitive performance on various prediction metrics.
• The model effectively learns polynomial combinations of protein features, enhancing prediction accuracy.
• Successfully assigned functions to proteins with very low sequence similarity and rare Gene Ontology terms.

Conclusions:

• GO-LTR offers a powerful approach for automatic protein function prediction.
• The model's ability to capture complex feature interactions improves accuracy, especially in challenging cases.
• Provides a valuable tool for annotating vast numbers of uncharacterized proteins.