Parameter Identifiability of Ligand Binding Models

Understanding the interactions of proteins with their ligands requires knowledge of molecular properties, such as binding site affinities and the effects that binding at one site exerts on binding at other sites (cooperativity). These properties cannot be measured directly and are usually estimated by fitting binding data with models that contain these quantities as parameters.

The Aldrich lab is greatly interested in mechanistic models of binding, but acknowledge that many commonly used binding measurements and models do not yield identifiable parameters. Therefore, we are prioritizing an investigation into why this type of analysis fails. The Aldrich lab presents a general method for answering the critical question of whether these parameters are identifiable (i.e., whether their estimates are accurate and unique). In cases in which parameter estimates are not unique, our analysis provides insight into the fundamental causes of nonidentifiability. By establishing a sound theoretical foundation, this approach can thus serve as a guide for the proper design and analysis of protein-ligand binding experiments for the Aldrich lab as well as other research groups.

Lead by Dr. Thomas Middendorf, this project is presented in two back-to-back JGP papers, on Structural Identifiability Middendorf and Aldrich (2017) and Practical Identifiability Middendorf and Aldrich (2017).  This work lays out the theoretical framework to design powerful binding measurements with which we can learn quantitative and qualitative information about how ligand-binding in multi-site systems. Currently, Dr. Middendorf is working on the remaining parts in the series, performing similar rigorous analysis to understand what we can learn from site-specific binding measurements and other more sophisticated approaches.

Log-error surface in the two-parameter space of a two-site sequential binding model with respect to noiseless binding data. Hines, Middendorf and Aldrich (2014).