Recurrent neural network based hybrid model for reconstructing gene regulatory network.

Area of Science:

• Systems Biology
• Computational Biology
• Genomics

Background:

• Gene regulatory networks (GRNs) are crucial for understanding how genomes control biological systems.
• Reconstructing GRNs is vital for applications in disease pathway identification, drug targeting, and diagnostics.
• Current methods for GRN reconstruction face challenges with biological complexity and data noise.

Purpose of the Study:

• To develop an advanced computational model for accurate GRN reconstruction.
• To leverage recurrent neural networks (RNNs) for modeling complex gene interactions.
• To enhance model performance and noise tolerance using a generalized extended Kalman filter.

Main Methods:

• Proposed a hybrid model combining Recurrent Neural Networks (RNNs) with a Generalized Extended Kalman Filter.
• Utilized backpropagation through time for training the RNN model.
• Tested the model on benchmark networks including DNA SOS repair, IRMA, and synthetic datasets.

Main Results:

• The developed RNN-based model accurately reconstructs GRNs, outperforming existing state-of-the-art methods.
• The model effectively captures non-linear and dynamic relationships within gene expression data.
• Demonstrated significant noise tolerance, with minimal performance degradation when 5% Gaussian noise was introduced.

Conclusions:

• The proposed hybrid RNN-Kalman filter model offers a superior approach for GRN reconstruction.
• The model's ability to handle noisy data makes it suitable for real-world biological applications.
• This work advances systems biology by providing a robust tool for deciphering gene regulatory mechanisms.