Semi-supervised multi-label collective classification ensemble for functional genomics.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in genomics

Background:

• Manual protein function prediction is cumbersome due to large-scale omics data.
• Existing computational methods often require extensive labeled data, which is difficult to obtain, especially for multi-label problems.
• Limited labeled data hinders accurate protein annotation.

Purpose of the Study:

• To develop an automated method for predicting protein functional properties.
• To address the challenge of protein function prediction with limited labeled data.
• To leverage diverse data sources including interaction networks and attribute features.

Main Methods:

• Framed protein function prediction as a semi-supervised multi-label collective classification (SMCC) problem.
• Proposed a novel generative model-based SMCC algorithm (GM-SMCC).
• Extended the method into an ensemble approach (EGM-SMCC) using multiple heterogeneous networks.

Main Results:

• GM-SMCC effectively computes label probability distributions for unannotated proteins.
• EGM-SMCC enhances prediction performance by propagating knowledge across heterogeneous networks.
• The proposed methods demonstrated effectiveness in predicting yeast protein functions and localization.

Conclusions:

• The developed GM-SMCC and EGM-SMCC algorithms are effective for protein function prediction with limited supervision.
• The proposed methods significantly outperform existing algorithms in comparative studies.
• This work provides a robust computational solution for protein annotation challenges.