Multi-objective Simulated Annealing Variants to Infer Gene Regulatory Network: A Comparative Study

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Gene Regulatory Networks (GRNs) govern cellular processes through gene interactions.
• Understanding GRN dynamics is crucial for deciphering gene expression regulation.
• Recurrent Neural Networks (RNNs) offer a promising approach for modeling GRNs.

Purpose of the Study:

• To computationally infer Gene Regulatory Networks (GRNs).
• To evaluate novel algorithms for parameter learning in RNN-based GRN models.
• To compare the performance of proposed algorithms against existing methods.

Main Methods:

• Application of a bi-objective RNN formulation for GRN modeling.
• Development and application of four algorithms based on Archived Multi Objective Simulated Annealing (AMOSA): AMOSAR, AMOFSA, AMOTSA, and AMOFTSA.
• Parameter learning using four gene expression time series datasets.

Main Results:

• Comparative performance analysis of the four proposed algorithms.
• Evaluation based on the number of GRNs in the non-dominated front, recall, precision, and F1 score.
• AMOFSA and AMOTSA demonstrated competitive performance.

Conclusions:

• Modified AMOSA algorithms (AMOFSA, AMOTSA) outperform AMOSAR and state-of-the-art methods.
• The proposed algorithms enhance the accuracy of GRN inference.
• This work contributes to a better understanding of gene expression regulation.