Sparse Zero-Sum Games as Stable Functional Feature Selection.

Area of Science:

• Systems Biology
• Computational Biology
• Machine Learning

Background:

• Large-scale systems biology applications involve complex datasets with hidden functional categories.
• Identical predictive power across features complicates traditional feature selection.
• Effective feature selection can reduce dimensionality and uncover functional insights.

Purpose of the Study:

• To propose a stable functional feature selection framework for systems biology.
• To leverage sparse zero-sum games for identifying relevant feature subsets.
• To reveal knowledge about functional classes of variables.

Main Methods:

• A sparse zero-sum game framework is developed.
• Feature subsets are ranked using a thresholding stochastic bandit.
• Theoretical analysis of the algorithm is provided.

Main Results:

• The proposed method demonstrates stable functional feature selection.
• Experiments on synthetic and real data show competitive performance.
• The approach is effective in both predictive accuracy and stability.

Conclusions:

• The sparse zero-sum game framework offers a robust solution for feature selection in systems biology.
• The method enhances understanding of functional variable classes.
• This approach is valuable for analyzing complex biological data.