Cancer survival analysis using semi-supervised learning method based on Cox and AFT models with L1/2 regularization.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate cancer diagnosis and prognosis rely on gene expression profiles.
• Cox (Cox proportional hazards) and AFT (accelerated failure time) models are standard for risk classification and survival prediction.
• Limitations include small sample sizes, censored data, and molecular heterogeneity confounding predictions.

Purpose of the Study:

• To develop a novel semi-supervised learning method to overcome limitations in Cox and AFT models for cancer survival analysis.
• To improve the accuracy of predicting patient treatment risk and survival time using gene expression data.

Main Methods:

• Proposed a semi-supervised learning framework integrating Cox and AFT models.
• Employed L1/2 regularization for efficient selection of relevant genes associated with disease.
• Applied the method to analyze real microarray gene expression and artificial datasets.

Main Results:

• The semi-supervised learning model significantly enhanced the predictive performance of both Cox and AFT models in survival analysis.
• Demonstrated successful application on four real-world microarray gene expression datasets.
• Validated improved predictive accuracy and risk classification capabilities.

Conclusions:

• The proposed method effectively increases usable training samples from censored data.
• Achieved high accuracy in identifying survival risk classes (Cox model) and predicting survival time (AFT model).
• Showcased strong capability in relevant biomarker selection, making it a suitable tool for clinical cancer research.