Can we Automatically Detect Biologically Relevant Order Parameters in Molecular Simulation? Comparing Long Timescale Simulations of Multiple Kinases

Abstract

Kinases are a ubiquitous component of important signaling pathways, including those frequently involved in cancer. Understanding the conformational heterogeneity of kinases has been shown to be important for understanding their function and their role as therapeutic targets. While thousands of structures are available via the PDB of kinases in various states, understanding how these states are kinetically and energetically related is key to making sense of these structures. Molecular dynamics has the power to give us this insight, however in simulating kinases structures from hundreds of states found in the PDB or from a single starting structure, we have found that defining order parameters to understand these dynamics can be far from straightforward. Analyzing our simulations, we optimized our order parameters to reflect the slowest processes, the starting structure of the simulation, or known kinase order parameters from the literature. To better facilitate this comparison, we have built a Python library (kinalysis) that can map any kinase structure or set of structures onto previously established kinase states (defined by DFG flip, C-helix switch, activation loop conformation, etc.). This can then be used to understand how well our generalizable methods for defining metastable states captures biologically relevant conformational changes. There is a careful balance between biasing an analysis with what you expect to see and allowing a simulation to tell you something you didn’t know before. In the long term, we hope this study informs how generalized methods of defining simulation-driven coordinates rather than coordinates hand-picked for a specific system add to our understanding of protein conformational dynamics.