GLIMS: A two-stage gradual-learning method for cancer genes prediction using multi-omics data and co-splicing network

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate identification of cancer genes is crucial for developing effective cancer diagnostics and therapeutics.
• Current methods for cancer gene identification face challenges due to the complexity of cancer and limited knowledge, necessitating improved approaches.
• Existing methods often struggle with accuracy when relying on limited omics data.

Purpose of the Study:

• To develop an advanced computational strategy for precise cancer gene identification.
• To enhance the accuracy of predicting cancer genes by integrating diverse biological data.
• To provide a robust tool for advancing cancer research and analysis.

Main Methods:

• A two-stage gradual-learning strategy named GLIMS was developed.
• The first stage employs a semi-supervised hierarchical graph neural network integrating multi-omics data and protein-protein interaction (PPI) networks.
• The second stage utilizes an unsupervised approach to refine predictions by incorporating co-splicing networks involved in post-transcriptional regulation.

Main Results:

• GLIMS demonstrated superior performance compared to existing state-of-the-art methods in identifying cancer genes.
• Systematic experiments on multi-omics cancer data validated the effectiveness of the GLIMS strategy.
• The method successfully integrated features from multi-omics data and biological networks.

Conclusions:

• GLIMS offers a significant advancement in cancer gene identification accuracy.
• The proposed method can serve as a valuable tool for researchers in cancer genomics and precision medicine.
• Integrating multi-omics data and network information, including post-transcriptional regulation, is key to improving cancer gene prediction.