Gene regulation network inference using k-nearest neighbor-based mutual information estimation: revisiting an old...

Area of Science:

• Systems Biology
• Computational Biology
• Bioinformatics

Background:

• Gene regulatory networks (GRNs) control cellular responses to internal and external cues.
• Reconstructing GRN topology from gene expression data is crucial for understanding cellular mechanisms and potential therapeutic benefits.
• Mutual Information (MI) is a key metric for inferring GRNs, but its application to continuous data presents challenges.

Purpose of the Study:

• To improve the accuracy of mutual information estimation for continuous gene expression data.
• To enhance the performance of gene regulatory network reconstruction algorithms.
• To introduce a novel, more effective GRN inference method.

Main Methods:

• Utilized k-nearest neighbor (kNN) mutual information (MI) estimation, specifically the Kraskov-Stoögbauer-Grassberger (KSG) algorithm, for enhanced accuracy with continuous data.
• Integrated the KSG-MI estimator with established GRN inference algorithms like Context Likelihood of Relatedness (CLR).
• Developed a new inference algorithm, Conditional Mutual Information Augmentation (CMIA), inspired by CLR and combined with KSG-MI.

Main Results:

• kNN-based MI estimation demonstrated significant error reduction for Gaussian distributions compared to fixed binning methods.
• The KSG-MI estimator significantly improved GRN reconstruction accuracy when used with algorithms like CLR.
• The novel CMIA algorithm, coupled with the KSG-MI estimator, outperformed existing methods in in-silico benchmarking.

Conclusions:

• The combined CMIA and KSG-MI method achieved a 20-35% improvement in precision-recall measures for GRN reconstruction on benchmark datasets.
• This advanced method offers a more reliable tool for discovering novel gene interactions.
• The findings facilitate better selection of gene candidates for experimental validation in biological research.