Developing an Improved Survival Prediction Model for Disease Prognosis.

Area of Science:

• Genetics and Molecular Biology
• Computational Biology
• Bioinformatics

Background:

• Machine learning is crucial in genetics and molecular biology for clinical research.
• Survival analysis models are vital for evaluating cancer treatment efficacy.
• High-dimensional genomic data presents a significant challenge for predictive modeling accuracy.

Purpose of the Study:

• To develop an improved survival prediction model for cancer.
• To address the limitations of high-dimensional genomic data in survival analysis.
• To enhance the prediction performance of survival learning models.

Main Methods:

• Proposed an improved survival prediction model integrating deep forest and self-supervised learning.
• Utilized a deep survival forest for adaptive learning of high-dimensional genomic data.
• Employed self-supervised learning to leverage unlabeled samples for performance enhancement.

Main Results:

• The proposed model demonstrated superior performance compared to four advanced survival analysis methods.
• Achieved improved prediction accuracy, evidenced by higher C-index and lower Brier scores.
• Validated on four cancer datasets from The Cancer Genome Atlas (TCGA).

Conclusions:

• The developed model effectively handles high-dimensional genomic data for robust survival prediction.
• Self-supervised learning significantly boosts model performance by utilizing unlabeled data.
• This computational tool can aid clinicians in personalizing cancer treatment strategies based on patient genomic characteristics and survival outcomes.