BiGSM: Bayesian inference of gene regulatory network via sparse modelling

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Gene regulatory network (GRN) inference is complex due to sparse matrices and noisy expression data, leading to prediction inaccuracies.
• Existing methods often provide fixed estimates, limiting comprehensive network analysis and confidence assessment.
• A Bayesian approach offers probabilistic link selection and statistical confidence insights, crucial for robust GRN inference.

Purpose of the Study:

• To develop a robust Bayesian method for gene regulatory network inference that handles data sparsity and noise.
• To create a method that provides probabilistic link selection and quantifies confidence in predicted GRN links.
• To rigorously benchmark the proposed method against state-of-the-art GRN inference techniques.

Main Methods:

• Proposed Bayesian inference of GRN via Sparse Modelling (BiGSM).
• Utilized maximum likelihood-based learning to infer posterior distributions of GRN links from noisy expression data.
• Leveraged GRN matrix sparsity for improved inference accuracy.

Main Results:

• BiGSM demonstrated superior performance in accuracy and robustness across various noise levels and data models in benchmark tests.
• Outperformed state-of-the-art methods like GENIE3, LASSO, LSCON, and Zscore using point-estimate measures.
• BiGSM uniquely provides posterior probabilities for GRN weights, enabling confidence assessment for each predicted link.

Conclusions:

• BiGSM offers a robust and informative approach to gene regulatory network inference.
• The method's ability to provide probabilistic outputs enhances the reliability and interpretability of inferred GRNs.
• BiGSM represents a significant advancement in computational methods for systems biology research.