Collaborative Research: CDS&E: Elucidating Binding using Bayesian Inference to Integrate Multiple Data Sources

点击翻译按钮获取中文摘要

With support from the Chemical Measurement and Imaging Program in the Division of Chemistry, Professors David Minh and John Chodera, and their groups at (respectively) the Illinois Institute of Technology and the Sloan Kettering Institute for Cancer Research, are developing statistical methods to study binding interactions between molecules. These interactions play critical roles in biology and materials technology. Full understanding of binding interactions can require integrating large amounts of data collected using multiple analytical instruments and experimental protocols. Existing statistical methods and software do not fully integrate data from multiple sources to produce useful knowledge. The Minh/Chodera team is pioneering the use of a new approach (a "Bayesian network") as a general framework for analyzing chemical measurement data from multiple instruments and protocols and for designing new experiments. The framework is usable for both small laboratory experiments and the massive datasets generated by automated instrumentation. The software (including a straightforward user interface) is utilized to teach the underlying principles in related courses, and will be made freely available online, along with tutorials and clear documentation. The Minh/Chodera team is developing chemometric methods and software for analyzing data related to binding. They are working to fuse data from diverse methods, including isothermal titration calorimetry (ITC), surface plasmon resonance (SPR), absorbance, fluorescence, and X-ray solution scattering. Key features of the software include automated parameter determination for physical binding models, and uncertainty propagation and quantification for model parameters. The research team also incorporates automated and principled model selection and hypothesis testing, and Bayesian experimental design to maximize acquisition of new information while minimizing cost. The software automatically constructs Bayesian networks that consider all sources of experimental error (e.g. dispensing, weighing, transfer, and measurement) for any experiment described by the Autoprotocol machine-readable standard. The software then performs Bayesian inference to weigh evidence for competing physical models, obtain credible intervals for thermodynamic and kinetic parameters, and propose new experiments. Robotic experiments, statistical inference, and Bayesian experimental design can be efficiently iterated to reduce model ambiguity and improve parameter precision. The team is using the software to advance knowledge of cooperativity between binding sites. A test application focuses on physiochemical properties that dictate site affinities and selectivities in human serum albumin.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.