Importance-Penalized Joint Graphical Lasso (IPJGL): differential network inference via GGMs

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Differential network inference aims to uncover gene interactions under varying conditions.
• Existing methods often overlook gene importance, potentially biasing results.
• Vital genes have lower mutation tolerance, necessitating consideration of their significance.

Purpose of the Study:

• To develop a novel differential network inference method that incorporates gene importance.
• To introduce a new metric (APC2) for evaluating differential gene pair levels.
• To identify potential cancer genes using the developed method on real-world datasets.

Main Methods:

• Developed the Importance-Penalized Joint Graphical Lasso (IPJGL) method based on Gaussian graphical models with adaptive gene importance regularization.
• Validated IPJGL using simulation experiments and real biological datasets.
• Proposed the APC2 metric for precise evaluation of differential network inference results.

Main Results:

• IPJGL successfully infers differential gene networks by considering gene importance.
• Application to TCGA colorectal and breast cancer datasets identified significant candidate cancer genes (SOST, RBBP8).
• Further analysis using the Reactome database confirmed the method's utility and biological relevance.

Conclusions:

• The Importance-Penalized Joint Graphical Lasso (IPJGL) method offers an improved approach to differential network inference by integrating gene importance.
• The identified candidate genes (SOST, RBBP8) warrant further investigation for their roles in colorectal and breast cancer.
• The proposed APC2 metric provides a valuable tool for assessing the accuracy of differential network inference.