Sparse Inverse Covariance Estimation with L0 Penalty for Network Construction with Omics Data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene coexpression and association networks are vital for understanding biological mechanisms.
• Gaussian graphical models with L1 penalty (LASSO) are common for high-dimensional omics data but have limitations like biased estimators and inconsistent model selection.
• Previous work on L0 regularized regression provides a foundation for improved network inference.

Purpose of the Study:

• To propose a robust and consistent method for learning gene network structures from high-dimensional omics data.
• To develop novel procedures for network construction and high-order gene-gene interaction detection.
• To address the limitations of L1-based methods in accurately identifying true biological networks.

Main Methods:

• Development of L0 regularized sparse inverse covariance estimation (L0RICE).
• Utilizing an efficient alternating direction (AD) method for structure learning.
• Application to omics data, including next-generation sequencing data.

Main Results:

• L0RICE demonstrates robustness and possesses the oracle property.
• The method successfully identifies high-order gene-gene correlation structures.
• Comparative analysis shows L0RICE significantly outperforms graphical lasso with a lower false positive rate.

Conclusions:

• L0 regularized structure learning offers a superior approach for omics data analysis.
• L0RICE provides a more accurate and reliable method for gene network construction and interaction detection.
• The proposed method advances the field of bioinformatics and computational biology for biological mechanism discovery.