GAPORE: Boolean network inference using a genetic algorithm with novel polynomial representation and encoding scheme

Inferring Boolean networks is crucial for modeling and analyzing gene regulatory networks from a systematic perspective. However, the state-of-the-art algorithms cannot accurately infer the topology and dynamics of Boolean networks due to the lack of an efficient approach to representing the unknown Boolean functions and the over-fit problem caused by the noise in time-series data. To address these problems, we propose a novel inference algorithm using a genetic algorithm with novel polynomial representation and encoding scheme (GAPORE) to reconstruct large-scale Boolean networks accurately. First of all, a novel symbolic polynomial representation method is introduced to efficiently represent the unknown Boolean functions of the candidate Boolean network as the symbolic polynomial dynamical equations. Then, a novel encoding scheme is developed to flexibly encode the symbolic polynomial dynamical equations by varying the effective lengths of the chromosomes. To reduce the over-fit problem, the l2-norm regularization is designed into the fitness evaluation in view of the network sparsity. In addition, the local search strategy is embedded into the hybrid genetic algorithm framework to strengthen the search capability. Extensive experiments demonstrate that GAPORE can infer the large-scale Boolean networks more accurately than state-of-the-art algorithms from the noisy time-series data.