Mixed-integer nonlinear optimisation approach to coarse-graining biochemical networks.

Area of Science:

• Biochemistry
• Computational Biology
• Systems Biology

Background:

• Quantitative modeling of biochemical networks is hindered by complexity, nonlinearity, and sparse parameter data.
• Detailed models are often computationally intractable and difficult to parameterize accurately.
• Coarse-grained models (CGMs) offer a simplified yet functionally relevant approach for analyzing biological systems.

Purpose of the Study:

• To develop a methodology for constructing effective coarse-grained models (CGMs) from detailed biochemical models.
• To simultaneously determine the structure and parameters of CGMs, especially when data is limited.
• To apply the methodology to a specific biological module and validate its predictive power.

Main Methods:

• A mixed-integer nonlinear-programming (MINLP) based optimization approach was employed.
• A genetic algorithm was used to simultaneously identify parameter values and prune unimportant reactions.
• The methodology was exemplified using the GTPase-cycle module involving M1 muscarinic acetylcholine receptor, Gq, and regulator of G protein signaling 4 (RGS4).

Main Results:

• Two CGMs with significantly reduced complexity (17 reactions from an initial 48) were developed.
• The resulting CGMs accurately fit experimental data and predict key signaling behaviors, including four limiting signaling regimes.
• The optimized CGM revealed an unconventional mechanism that explains the biphasic response observed with certain drugs.

Conclusions:

• The MINLP-based optimization approach provides an effective strategy for coarse-graining complex biochemical networks.
• Coarse-graining not only reduces model order but also yields mechanistic insights.
• This methodology is broadly applicable to larger and more complex biochemical systems, aiding in pharmacological understanding.