A disease-related gene mining method based on weakly supervised learning model.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Predicting disease-related genes aids in understanding disease pathology and molecular mechanisms.
• Traditional gene screening methods struggle with datasets containing weak label information and few known disease genes.

Purpose of the Study:

• To develop an effective disease-related gene mining method.
• To address limitations of traditional methods in handling weak labels and sparse known gene data.

Main Methods:

• A two-step weakly supervised learning model was designed.
• Step 1: Differentially expressed genes were screened using a weakly supervised model that leverages strong and weak label information.
• Step 2: Disease-related genes were identified from the differentially expressed set using transductive support vector machine with a difference kernel function.

Main Results:

• The weakly supervised learning model identified a stable and complete set of differentially expressed genes.
• The difference kernel function enabled more accurate evaluation of gene relationships in a difference space.
• The method effectively utilized known disease-related gene information.

Conclusions:

• The developed disease-related gene mining method significantly improves prediction precision.
• The approach outperforms existing methods in identifying disease-related genes, especially with limited data.