A Bi-Objective RNN Model to Reconstruct Gene Regulatory Network: A Modified Multi-Objective Simulated Annealing...

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Gene Regulatory Networks (GRNs) are crucial for understanding cellular processes and gene expression.
• Computational reconstruction of GRNs from gene expression data is a key research area.
• Recurrent Neural Networks (RNNs) offer a powerful modeling approach for GRN reconstruction.

Purpose of the Study:

• To modify the RNN formulation for GRN reconstruction by incorporating a new objective function, creating a bi-objective model.
• To develop and apply a novel algorithm, AMOSA-GRN, for improved GRN reconstruction.

Main Methods:

• Modified an existing multi-objective meta-heuristic algorithm, Archived Multi Objective Simulated Annealing (AMOSA), to handle the bi-objective RNN formulation.
• Applied the modified algorithm (AMOSA-GRN) to gene expression datasets to obtain a collection of non-dominated GRNs (Archive).
• Utilized ensemble averaging on the obtained archives from multiple AMOSA-GRN executions.

Main Results:

• The AMOSA-GRN algorithm successfully generated a collection of non-dominated GRNs.
• Ensemble averaging of the GRN archives resulted in improved reconstruction accuracy.
• The accuracy of the reconstructed GRNs, compared to a gold standard, ranged from 87.5% to 100%.

Conclusions:

• The bi-objective RNN formulation and the AMOSA-GRN algorithm represent a significant advancement in computational GRN reconstruction.
• The method demonstrates high accuracy and effectiveness in inferring gene regulatory relationships from expression data.
• This approach holds promise for deeper insights into cellular mechanisms and gene regulation.