A new computational method to split large biochemical networks into coherent subnets.

Area of Science:

• Systems biology
• Biochemical network analysis
• Computational biology

Background:

• Biochemical networks possess unique characteristics, including sparse connectivity with some highly connected nodes.
• Metabolite nodes are classified as internal (with mass balance) or external (without).
• Reclassifying internal nodes as external can simplify complex metabolic networks into subnetworks.

Purpose of the Study:

• To introduce Netsplitter, a novel method for controlled partitioning of complex metabolic networks.
• To improve upon existing node connectivity-based partitioning methods.
• To provide users with interactive control over subnetwork division.

Main Methods:

• Netsplitter combines local connection degree partitioning with global connectivity from random walks.
• The method employs progressive partitioning with interactive visual matrix presentation.
• Special strategies are integrated to maintain network integrity and minimize information loss.

Main Results:

• Partitioning a genome-scale *Arabidopsis thaliana* network yielded 10 medium-sized subnetworks, encompassing 66% of the network.
• The flavonoid subnetwork was naturally separated into four functionally distinct subnets.
• The quantitative measure 'efficacy' demonstrated improved partitioning across bacterial, plant, and mammal metabolic networks.

Conclusions:

• Netsplitter offers a significant improvement over connection degree partitioning, balancing subnet sizes and minimizing mass balance constraint removal.
• The method provides interactive user control over node selection and partitioning granularity.
• The blocking transformation offers a powerful visualization tool for network exploration.