Enhanced protein fold recognition through a novel data integration approach.

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• Protein fold recognition is crucial for determining 3D protein structures.
• Multiple data sources, including sequence alignments and structural properties, are used for fold discrimination.
• Combining these diverse data sources efficiently is a significant challenge in protein classification.

Purpose of the Study:

• To develop a novel kernel-based approach for integrating multiple heterogeneous data sources for protein fold recognition.
• To propose information-theoretic methods based on Kullback-Leibler (KL) divergence for data integration.
• To evaluate the effectiveness of these methods in improving protein fold classification accuracy.

Main Methods:

• Utilized a kernel-based approach integrating multiple data sources.
• Developed two formulations, MKLdiv-dc and MKLdiv-conv, based on KL divergence between input and output kernel matrices.
• Employed difference of convex (DC) programming for MKLdiv-dc and projected gradient descent for MKLdiv-conv.

Main Results:

• Achieved state-of-the-art performance on the SCOP PDB-40D benchmark dataset for protein fold prediction.
• MKLdiv-dc improved fold discrimination accuracy to 75.19%, a >5% increase over existing methods.
• Demonstrated competitive performance on a yeast protein function prediction task.

Conclusions:

• The proposed MKLdiv-dc and MKLdiv-conv methods effectively integrate diverse data sources for protein fold recognition.
• These methods offer insights into the relative importance of different data sources.
• Achieved superior performance in protein fold prediction and competitive results in protein function prediction.