How to infer gene networks from expression profiles, revisited

Area of Science:

• Systems Biology
• Computational Biology
• Genomics

Background:

• Gene regulatory network (GRN) inference is crucial for understanding complex biological processes.
• Existing methods include information theory, Bayesian networks (BNs), dynamic Bayesian networks (DBNs), and ordinary differential equations (ODEs).
• Previous assessments of these methods used various in silico and in vivo datasets.

Purpose of the Study:

• To evaluate the performance of recent network inference algorithms.
• To introduce and assess a novel non-parametric learning approach based on nonlinear dynamical systems.
• To compare the accuracy and computational efficiency of different GRN inference methods.

Main Methods:

• Assessed performance of novel non-parametric learning algorithms based on nonlinear dynamical systems.
• Compared these with traditional approaches like DBNs, Granger causality, and ODE models.
• Utilized in silico and in vivo datasets for evaluation.

Main Results:

• Non-parametric approaches provide higher accuracy for large GRNs (hundreds of genes) but incur significant computational costs.
• For smaller systems, DBNs demonstrate comparable accuracy and computational efficiency to non-parametric methods.
• Both non-parametric and DBN approaches outperform Granger causality and simple ODE models.

Conclusions:

• Novel non-parametric methods offer superior accuracy for large-scale GRN inference.
• DBNs present a computationally efficient and accurate alternative for smaller gene networks.
• The choice of GRN inference method depends on network size and desired balance between accuracy and computational resources.