EXAMINE: a computational approach to reconstructing gene regulatory networks

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Linear models for gene network reverse-engineering often face underdetermined systems due to excessive parameters.
• The practical utility of linear models in this context remains unclear.

Purpose of the Study:

• To develop an improved method, EXpression Array MINing Engine (EXAMINE), for inferring gene regulatory networks from time-series gene expression data.
• To address the excessive-parameter problem and clarify the utility of linear models in gene network inference.

Main Methods:

• EXAMINE utilizes sparse graph theory and an adaptive-connectivity model to manage excessive parameters.
• An incremental adaptive fitting process ensures the discovery of the most parsimonious network structure.
• The method reduces parameters by O(N) compared to fully connected models, enhancing regulatory network recovery.

Main Results:

• EXAMINE efficiently generates regulatory networks from time-series gene expression data.
• A systematic study provides guidelines for linear model application: effective for small systems (3-20 elements), requiring clustering or additional data for large systems.
• Application to rat central nervous system development data (112 genes) successfully generated statistically significant regulatory pathways.

Conclusions:

• EXAMINE offers an effective solution for inferring gene regulatory networks, particularly by addressing parameter challenges in linear models.
• The study provides practical guidelines for applying linear models in gene network analysis based on system scale.
• EXAMINE, especially when combined with clustering, demonstrates efficacy in analyzing complex biological systems like neural development.