Incorporating existing network information into gene network inference.

Area of Science:

• Systems Biology
• Computational Biology
• Genomics

Background:

• Gene regulatory network inference from gene expression data is crucial for understanding cellular processes.
• Ordinary differential equations (ODEs) provide a robust framework for modeling these networks.
• Integrating diverse data types, such as ChIP-seq, can enhance network inference accuracy.

Purpose of the Study:

• To extend ODE-based gene network inference to incorporate prior biological network information.
• To develop a general optimization framework that utilizes a priori network data and promotes network sparsity.
• To validate the method's performance and theoretical convergence properties.

Main Methods:

• Developed a generalized optimization framework extending ODE methodology.
• Incorporated prior network information (e.g., from ChIP-chip/ChIP-seq) as constraints.
• Utilized regularization parameters to encourage network sparsity.
• Provided theoretical convergence proofs and probabilistic interpretations.

Main Results:

• Demonstrated improved performance on simulated data when incorporating prior network information.
• Showed the method overcomes limitations of sparse observational data.
• Validated effectiveness even with partially incorrect prior network information.
• Applied successfully to reconstruct the core regulatory network of embryonic stem cells.

Conclusions:

• Integrating prior network information significantly improves the accuracy of ODE-based gene network inference.
• The proposed framework is robust, handling imperfect prior data and sparse observations.
• This approach leads to more accurate representations of biological regulatory networks.