Finding gene network topologies for given biological function with recurrent neural network.

Area of Science:

• Systems Biology
• Computational Biology
• Bioinformatics

Background:

• Identifying biochemical networks for specific functions is crucial but challenging in systems biology.
• Exhaustive search methods are limited to small networks and simple functions, failing to scale for complex biological systems.

Purpose of the Study:

• To develop a scalable computational method for discovering complex regulatory network topologies and functions.
• To overcome the limitations of exhaustive search methods in systems biology.

Main Methods:

• Training a recurrent neural network (RNN) to perform a desired biological function.
• Employing a systematic perturbative method to interrogate trained RNNs and extract the underlying regulatory network.
• Validating discovered networks using realistic biological response functions like the Hill-function.

Main Results:

• Successfully trained RNNs to perform complex biological functions.
• Distilled underlying regulatory network logic and topology from trained RNNs.
• Demonstrated that RNN-derived networks can achieve biological functions with realistic response functions.

Conclusions:

• The developed RNN-based method provides a scalable approach to link network topology and function in systems biology.
• This approach aids in uncovering complex regulatory logic and network structures for intricate biological tasks.
• The findings offer a new computational tool for advancing systems biology research.