Large-Scale Recurrent Neural Network Based Modelling of Gene Regulatory Network Using Cuckoo Search-Flower...

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Accurate prediction of genetic networks is a key challenge in the postgenomic era.
• Recurrent Neural Networks (RNNs) are effective for small-scale genetic network modeling but underperform on large-scale networks.
• Existing computational tools struggle with the complexity of large genetic regulatory networks.

Purpose of the Study:

• To develop an improved computational methodology for predicting large-scale genetic networks.
• To enhance the accuracy of inferring gene regulatory relationships.
• To address the limitations of standard Recurrent Neural Networks in modeling complex biological systems.

Main Methods:

• A hybrid approach combining Cuckoo Search (CS) and Flower Pollination Algorithm (FPA) with Recurrent Neural Networks (RNNs).
• CS is utilized to identify optimal regulator combinations.
• FPA is employed to fine-tune RNN model parameters for enhanced performance.

Main Results:

• The hybrid CS-FPA-RNN methodology demonstrated superior performance on a large-scale artificial genetic network, both with and without noise.
• The approach significantly increased the inference of correct genetic regulations while reducing false positives.
• Validation on the Escherichia coli DNA SOS repair network confirmed the method's effectiveness on real-world biological data.

Conclusions:

• The proposed hybrid optimization technique significantly improves the accuracy of large-scale genetic network inference compared to traditional RNNs.
• This methodology offers a robust solution for deciphering complex gene regulatory interactions in biological systems.
• While effective, the hybrid approach incurs increased computational time due to the complex optimization process.