Enhancing the Lasso Approach for Developing a Survival Prediction Model Based on Gene Expression Data.

Area of Science:

• Oncology
• Bioinformatics
• Statistical Genetics

Background:

• Gene expression data is crucial for developing cancer survival prediction models.
• The Least Absolute Shrinkage and Selection Operator (LASSO) is commonly used for gene selection but may miss true positive genes (false negatives).
• Cross-validation (CV) is often used to determine LASSO's tuning parameter, but this can lead to suboptimal gene identification.

Purpose of the Study:

• To develop a novel method for estimating the number of true positive (TP) genes in survival prediction models.
• To address the limitation of the LASSO method in identifying all relevant genes.
• To improve the accuracy and comprehensiveness of gene selection for cancer survival prediction.

Main Methods:

• Developed a method to estimate TP genes by assuming a mixture distribution for LASSO estimates.
• Performed simulation studies to evaluate the precision of the developed method.
• Applied the method to a real gene expression dataset to identify survival-correlated genes.

Main Results:

• The developed method precisely estimates the number of TP genes in simulation studies.
• The method successfully identified genes correlated with survival that were missed by traditional CV methods.
• This approach enhances the detection of relevant genes in oncology survival prediction.

Conclusions:

• The proposed method offers a more effective way to identify true positive genes for cancer survival prediction.
• It overcomes limitations of standard LASSO and CV approaches in gene selection.
• This advancement can lead to more accurate and robust survival models in oncology.