GAEM: Genetic Algorithm based Expectation-Maximization for inferring Gene Regulatory Networks from incomplete data.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Inferring Gene Regulatory Network (GRN) structures from incomplete gene expression data is a significant challenge in bioinformatics.
• Existing methods like the Path Consistency Algorithm based on Conditional Mutual Information (PCA-CMI) struggle with missing data, necessitating imputation techniques.

Purpose of the Study:

• To present the GAEM algorithm, a novel method for inferring GRN structures from incomplete gene expression datasets.
• To address the limitations of traditional imputation-then-GRN-inference approaches by iteratively updating both missing values and the GRN structure.

Main Methods:

• The GAEM algorithm combines a Genetic Algorithm and Expectation-Maximization in an iterative approach.
• It learns GRN structure by repeatedly updating imputed values based on the inferred GRN until convergence.

Main Results:

• GAEM was evaluated under various missingness percentages (5%, 15%, 40%) and mechanisms.
• Results on the DREAM3 dataset indicate GAEM is a reliable method, outperforming traditional approaches, particularly for smaller network sizes.

Conclusions:

• The GAEM algorithm offers a robust solution for GRN inference from incomplete gene expression data.
• The GAEM R package is available, facilitating its application in bioinformatics research.