Inference of biochemical network models in S-system using multiobjective optimization approach.

Area of Science:

• Systems Biology
• Computational Biology
• Biochemical Network Inference

Background:

• Inferring biochemical networks (gene regulatory, protein-protein interaction, metabolic) from time-course data is a key challenge in systems biology.
• Current methods often use penalty terms for kinetic orders, but tuning the penalty weight for model structure inference remains problematic.
• A lack of guidelines exists for selecting appropriate penalty weights to infer suitable biochemical network model structures.

Purpose of the Study:

• To develop an interactive inference algorithm for creating realizable S-system structures of biochemical networks.
• To formulate the network inference as a multiobjective optimization problem for simultaneous error and interaction minimization.
• To provide a method that bypasses the need for arbitrary penalty weight tuning.

Main Methods:

• An interactive inference algorithm is proposed for S-system structure inference.
• The problem is framed as a multiobjective optimization task, minimizing concentration error, slope error, and interaction measure.
• The epsilon-constraint method is employed to minimize interaction measure under concentration and slope error constraints.

Main Results:

• The algorithm successfully infers realizable S-system structures for biochemical networks.
• The multiobjective optimization approach effectively balances model accuracy and network complexity.
• The proposed method guarantees a minimum interaction network solution, avoiding subjective penalty weight selection.

Conclusions:

• The interactive inference algorithm offers a robust approach to biochemical network modeling.
• This method provides a systematic way to infer S-system models without relying on empirical penalty weight tuning.
• The findings contribute to advancing the quantitative understanding of biological systems through accurate network inference.