Multi-task Learning Graph Neural Networks for Cancer Prognosis Prediction with Genomic Data.

Area of Science:

• Computational biology
• Oncology
• Bioinformatics

Background:

• Accurate cancer prognosis prediction is crucial for precision oncology.
• Limited data samples pose a significant challenge for developing robust predictive models.
• Existing methods struggle to generalize well across different cancer types with sparse data.

Purpose of the Study:

• To develop a novel approach combining multi-task learning (MTL) and graph neural networks (GNNs) for improved cancer prognosis prediction.
• To address the challenge of predicting prognosis in cancers with limited data samples.
• To leverage shared biological information across different cancer types.

Main Methods:

• Gene-gene interactions were represented as a graph network.
• Multi-task learning (MTL) was employed to capture relationships between genes involved in oncogenesis and cancer progression.
• Graph neural networks (GNNs) were utilized to model these gene interactions.

Main Results:

• The proposed MTL and GNN approach significantly improved cancer prognosis prediction for cancers with limited data, such as colon adenocarcinoma.
• A 24% increase in the area under the precision-recall curve (AUPRC) was achieved by leveraging shared gene-gene interactions across cancer types.
• The model demonstrated enhanced predictive performance by learning from related cancer datasets.

Conclusions:

• The combination of MTL and GNNs offers a powerful strategy for enhancing cancer prognosis prediction, particularly in data-scarce scenarios.
• This approach effectively utilizes cross-cancer gene interaction data to improve precision oncology.
• The findings highlight the potential of integrating graph-based learning and MTL for advancing smart healthcare solutions in cancer research.