GeneSPIDER2: large scale GRN simulation and benchmarking with perturbed single-cell data

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Single-cell data is crucial for gene regulatory network (GRN) inference, with benchmarks relying on simulated data.
• Existing single-cell simulators lack the capability to model gene perturbations, a key biological process.
• Generating large-scale GRNs presents computational and stability challenges.

Purpose of the Study:

• To introduce GeneSPIDER2, an updated MATLAB toolbox for GRN benchmarking, inference, and analysis.
• To enhance the generation of large-scale GRNs with realistic topological properties.
• To enable simulation of single-cell data, including unique features for modeling genetic perturbations.

Main Methods:

• Developed improved software modules within GeneSPIDER2 for enhanced capabilities and performance.
• Implemented algorithms for generating large GRNs with scale-free degree distribution and modularity.
• Introduced a novel simulation module for generating single-cell data based on genetic perturbations.

Main Results:

• GeneSPIDER2 can generate large GRNs exhibiting biologically realistic topological features.
• The toolbox successfully simulates single-cell data, incorporating the effects of genetic perturbations.
• Simulated single-cell data demonstrated similar properties to real Perturb-seq data from two cell lines.

Conclusions:

• GeneSPIDER2 provides a robust platform for GRN inference and analysis using simulated single-cell data.
• The ability to simulate gene perturbations offers a unique advantage for benchmarking GRN inference methods.
• The validation against real data confirms the utility of GeneSPIDER2 for generating realistic synthetic single-cell datasets.