GD-Net: An Integrated Multimodal Information Model Based on Deep Learning for Cancer Outcome Prediction and...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multimodal data offers valuable insights for cancer prognosis and survival prediction.
• Integrating heterogeneous biological data presents computational challenges due to complex molecular interactions and limited sample sizes.

Purpose of the Study:

• To introduce GD-Net, a novel graph deep learning algorithm for enhanced cancer survival prediction.
• To leverage early fusion of multimodal information and an XGBoost module for feature extraction.

Main Methods:

• GD-Net employs graph deep learning for multimodal data integration.
• An interpretable XGBoost module is utilized for efficient feature extraction.
• The algorithm was applied to eight cancer datasets for performance evaluation.

Main Results:

• GD-Net achieved an average accuracy of 72% and outperformed benchmarking methods with a 7.9% higher C-index.
• Ablation experiments confirmed that multimodal integration significantly improves prediction accuracy over single-modality models.
• Case studies identified and validated key genes, miRNAs, and methylated genes as informative prognostic biomarkers.

Conclusions:

• GD-Net demonstrates superior accuracy and competitiveness in real-time cancer outcome prediction.
• The model is an effective tool for identifying novel multimodal prognosis biomarkers.
• Multimodal data integration is crucial for improving the accuracy of cancer survival prediction.