Assessment of network inference methods: how to cope with an underdetermined problem

Area of Science:

• Systems Biology
• Computational Biology
• Bioinformatics

Background:

• Biological network inference is crucial for understanding complex biological systems.
• Numerous network inference algorithms exist, necessitating robust performance assessment.
• Current assessments often overlook the underdetermined nature of network inference problems.

Purpose of the Study:

• To propose a novel procedure for assessing gene regulatory network (GRN) inference methods.
• To rigorously account for the underdetermined nature of the network inference problem.
• To introduce a revised confusion matrix definition excluding non-inferable gene regulations.

Main Methods:

• Developed a new assessment procedure for GRN inference methods.
• Incorporated causal inference to distinguish inferable from non-inferable gene regulations.
• Utilized a modified confusion matrix that excludes errors from non-inferable interactions.

Main Results:

• Applied the new procedure to the DREAM 4 In Silico Network Challenge dataset.
• Demonstrated that network inferability significantly impacts the performance ranking of inference methods.
• Observed a marked change in method rankings compared to traditional assessment approaches.

Conclusions:

• The proposed assessment procedure offers a more accurate evaluation of GRN inference methods.
• Accounting for network inferability is essential for reliable benchmarking in systems biology.
• This approach refines the comparison of algorithms in the growing field of biological network inference.