Bayesian Orthogonal Least Squares (BOLS) algorithm for reverse engineering of gene regulatory networks.

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Gene regulatory network (GRN) reverse engineering is computationally challenging with large gene numbers and limited experimental data.
• Linear system-based GRN reverse engineering is ill-posed, sensitive to noise, and requires robust optimization.
• Limited microarray data necessitates rigorous algorithms for accurate GRN elucidation.

Purpose of the Study:

• To develop a novel algorithm for reverse engineering gene regulatory networks using linear systems.
• To address the challenges of limited data points and noise sensitivity in GRN reconstruction.
• To improve the accuracy and efficiency of identifying true positive interactions in gene regulatory networks.

Main Methods:

• The proposed algorithm combines orthogonal least squares, second-order derivative for network pruning, and Bayesian model comparison.
• The gene regulatory network is decomposed into smaller, manageable unit networks.
• Each unit network is assigned a confidence level, P(D|Hi), to assess the accuracy of its elucidation.

Main Results:

• The algorithm successfully elucidates gene regulatory networks even with limited experimental data and noisy conditions.
• Evaluation on synthetic data shows performance is dependent on gene count, noise level, and data points.
• Analysis of Saccharomyces cerevisiae expression data revealed a significant number of known physical or genetic interactions.
• Comparison with Sparse Bayesian Learning demonstrated the proposed algorithm yields sparser solutions with fewer false positives.

Conclusions:

• The developed algorithm effectively elucidates gene regulatory networks using limited experimental data.
• The algorithm demonstrates robustness in handling noisy data, a common issue in biological experiments.
• Validation with yeast expression data confirms the reliable identification of known biological interactions.
• The algorithm shows superior performance compared to Sparse Bayesian Learning, particularly with noisy and limited datasets.