Quantitative comparison of alternative methods for coarse-graining biological networks

Area of Science:

• Computational biology
• Statistical mechanics
• Data science

Background:

• Markov models and master equations are vital for dynamic processes like protein conformational changes.
• Complex models present challenges due to numerous components and connections.
• Coarse-graining methods simplify these models for better understanding.

Purpose of the Study:

• To compare various coarse-graining methods for Markov models.
• To identify methods that produce models most faithful to original states.
• To evaluate performance using Bayesian model comparison.

Main Methods:

• Bayesian model comparison framework.
• Evaluation of coarse-graining algorithms: Bayesian agglomerative clustering engine, hierarchical Nyström expansion graph (HNEG), Perron cluster cluster analysis (PCCA), PCCA+, and most probable paths algorithm.
• Analysis of state population similarity and entropy.

Main Results:

• Bayesian agglomerative clustering engine and HNEG demonstrated superior performance.
• The original PCCA method outperformed PCCA+ and most probable paths algorithm.
• Performance differences were qualitatively significant, not minor shifts.
• Method performance correlated with the entropy of coarse-grainings.

Conclusions:

• Bayesian agglomerative clustering engine and HNEG are recommended for coarse-graining complex Markov models.
• The original PCCA remains a competitive method.
• Entropy of coarse-grained states is a useful indicator of method performance.
• Prioritizing similar state populations improves coarse-graining accuracy.