Learning Asynchronous Boolean Networks From Single-Cell Data Using Multiobjective Cooperative Genetic Programming.

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• High-throughput single-cell technologies generate vast gene expression data for computational modeling of gene regulatory networks (GRNs).
• Existing Boolean network (BN) models for GRNs often rely on bulk time-series data and assume synchronous updates, which is biologically unrealistic.
• The synchronous update assumption in BN modeling simplifies computation but limits biological accuracy.

Purpose of the Study:

• To propose a novel framework, SgpNet, for inferring asynchronous Boolean networks (BNs) from single-cell data.
• To address the limitations of synchronous update schemes in current GRN modeling.
• To develop a method that matches the asynchronous state transition graph (STG) from single-cell data while preserving GRN sparsity.

Main Methods:

• Formulated the inference of asynchronous BNs as a multiobjective optimization problem.
• Employed genetic programming to encode Boolean functions as trees and utilized cooperative coevolution for simultaneous network function evolution.
• Incorporated a regulator preselection strategy for enhanced learning efficiency and an error threshold estimation heuristic for parameter tuning.

Main Results:

• SgpNet achieved comparable inference accuracy to state-of-the-art methods on both synthetic and experimental single-cell data.
• The proposed framework significantly reduced the number of parameters compared to existing methods.
• SgpNet eliminated artificial restrictions on Boolean function structures and demonstrated potential for scalability through parallelization.

Conclusions:

• SgpNet provides a more biologically realistic approach to modeling GRNs by adopting an asynchronous update scheme.
• The framework offers improved efficiency and flexibility in inferring BNs from single-cell data.
• SgpNet represents a significant advancement in computational GRN modeling, enabling more accurate and scalable network inference.