Integrative gene network construction to analyze cancer recurrence using semi-supervised learning.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Cancer recurrence prediction is crucial but challenging due to small sample sizes and numerous genes.
• Existing supervised methods for cancer recurrence prediction are limited by their reliance on few labeled samples.
• Semi-supervised learning offers a promising alternative for analyzing complex genomic data in cancer research.

Purpose of the Study:

• To develop a novel semi-supervised learning algorithm for accurate cancer recurrence prediction.
• To leverage graph regularization and protein interaction data for enhanced gene expression analysis.
• To identify key cancer genes and their roles in recurrence through network analysis.

Main Methods:

• Transformed gene expression data into a graph structure for semi-supervised learning.
• Integrated protein interaction data with gene expression data to identify functionally related gene pairs.
• Applied a graph regularization approach to a constructed graph with labeled and unlabeled nodes for recurrence prediction.

Main Results:

• Achieved an average accuracy improvement of 24.9% across three cancer datasets compared to existing methods.
• Identified gene networks significantly associated with cancer recurrence-related biological functions through functional enrichment.
• Demonstrated the efficacy of the proposed semi-supervised learning algorithm in predicting cancer recurrence.

Conclusions:

• The novel semi-supervised learning algorithm significantly improves cancer recurrence prediction accuracy.
• The identified gene networks provide insights into the biological mechanisms underlying cancer recurrence.
• The developed C++ algorithm is freely available for Linux and MS Windows, facilitating further research.