EECFS: Efficient Ensemble Causal Feature Selection for High-Dimensional Molecular Data.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• High-dimensional biological data with limited samples pose prediction challenges.
• Causal feature selection offers improved interpretability and robustness over statistical methods.
• Existing constraint-based causal methods are computationally intensive, especially in spouse discovery.

Purpose of the Study:

• To develop an efficient ensemble causal feature selection algorithm (EECFS) to reduce computational cost.
• To apply causal feature selection to synonymous variant effect prediction, developing CFDPSM.
• To improve the efficiency and performance of biological prediction tasks.

Main Methods:

• Proposed EECFS, an ensemble causal feature selection algorithm with an efficient spouse discovery strategy.
• Evaluated EECFS on 16 Bayesian network and 17 real-world datasets.
• Developed CFDPSM for variant effect prediction, identifying Markov blanket features from multi-omics data.

Main Results:

• EECFS demonstrated improved efficiency and competitive/superior predictive performance against 11 methods.
• CFDPSM identified a compact set of 30 Markov blanket features from over 23,000 initial features.
• CFDPSM outperformed 13 existing variant effect prediction methods, offering enhanced interpretability.

Conclusions:

• EECFS provides an efficient and effective approach for causal feature selection in biological prediction.
• CFDPSM represents a significant advancement in synonymous variant effect prediction, balancing performance and interpretability.
• Causal feature selection holds great promise for advancing biological data analysis and interpretation.