Conclusions via unique predictions obtained despite unidentifiability--new definitions and a general method

Area of Science:

• Systems Biology
• Computational Biology
• Mathematical Biology

Background:

• Model-based data analysis is predicted to revolutionize biology.
• Hypothesis testing in systems biology often focuses on model rejection, but non-rejection can be a weak conclusion due to over-parameterization and parameter uncertainty.

Purpose of the Study:

• To formally define and analyze the concept of a "core prediction" in systems biology.
• To introduce a novel method for core prediction analysis that addresses parameter uncertainty and the curse of dimensionality.

Main Methods:

• Formal definition and analysis of "core predictions" as uniquely identified properties essential for model explanation.
• Development of a new method for core prediction analysis that characterizes acceptable parameter spaces in relevant directions.
• Comparison of the new method with profile likelihood for practical identifiability and generalization to observability.

Main Results:

• A "core prediction" is a robust conclusion, as strong as model rejection, even with non-unique parameter identification.
• The new method for core prediction analysis is efficient, avoiding the curse of dimensionality.
• The method demonstrates practical identifiability comparable to profile likelihood and generalizes it to observability.

Conclusions:

• Core predictions offer a stronger basis for conclusions in systems biology than simple non-rejection of models.
• The novel analysis method enhances confidence in systems biology findings by providing justified conclusions.
• This work distinguishes between strong conclusions and mere suggestions in model-based biological data analysis.