TSPLASSO: A Two-stage Prior LASSO Algorithm for Gene Selection using Omics Data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Feature selection is crucial for identifying cancer genes from omics data.
• Existing methods often overlook valuable prior knowledge about known cancer genes.
• Integrating prior cancer gene information can enhance the accuracy of feature selection.

Purpose of the Study:

• To propose a novel two-stage prior LASSO (TSPLASSO) method for cancer gene identification.
• To leverage prior knowledge of known cancer genes in the feature selection process.
• To simultaneously perform cancer gene selection and sample classification using omics data.

Main Methods:

• TSPLASSO employs a two-stage approach using LASSO regression.
• Stage one uses linear regression to select candidate genes correlated with prior cancer genes.
• Stage two uses logistic regression for final gene selection and sample classification.

Main Results:

• TSPLASSO demonstrated significant improvements (5%-400%) in variable selection accuracy across multiple datasets.
• The method showed robustness against data noise and variations in prior cancer gene information.
• TSPLASSO outperformed six state-of-the-art algorithms in accuracy and stability.

Conclusions:

• TSPLASSO offers an efficient, stable, and practical algorithm for cancer gene discovery from omics data.
• The method effectively integrates prior biological knowledge into feature selection.
• TSPLASSO advances biomedical and health informatics by improving the analysis of omics data.