(508f) Optimal Probes: A Machine Learning Platform for Design of Experimental Probes for Protein Dynamics | AIChE

(508f) Optimal Probes: A Machine Learning Platform for Design of Experimental Probes for Protein Dynamics


Shukla, D. - Presenter, University of Illinois at Urbana-Champaign
Mittal, S., UIUC
In this talk, we present our recent work focused on filling the void that exists in the study of membrane protein dynamics, by a fruitful combination of computation and experiment. Despite the ubiquitousness and the diversity in their function in the physical biology of the cell, study of membrane proteins has been limited owing to huge sizes and the difficulty to probe them in their native environment within the lipid bilayer. Membrane proteins play crucial roles in signal transduction, variety of substrate transport, enzymatic activity, immune response and are popular drug targets. Few crystal structures are available limiting the structural information, is mostly restricted to only few functionally diverse intermediates and metastable state that the protein adopts. We show how machine learning can be used to combine computational and biophysical experimental methodologies for studying membrane proteins, with the vision that in such a scenario the result is often greater than the sum of the parts. First, we present algorithms that use molecular dynamics (MD) simulation data, to predict the optimal reaction coordinates of an experimental technique, in order to perform experiments that are optimized to study membrane protein conformation dynamics. When there are multiple experiments on a target protein, it is useful if the they provide overlapping yet, distinct insights into it’s dynamics. Second, we determine which experimental dataset adds orthogonal information to what the theoretical/experimental studies already provide and update the model accordingly. Finally, we demonstrate a cloud-based GUI platform, which would act as a bridge between the available experimental and theoretical information; and provide the most appropriate future set of studies for a given protein of interest. The cloud platform is based on National Data Service's BROWN DOG methodology, which provides the web-based data access proxy (DAP) for file conversions and Data tilling service (Clowder) for metadata extraction. These services and tools aid in the curation, accessing and indexing of data as well as the software that might be leveraged for this purpose. Researchers would be able to use this resource to aid in the design of experiments in their studies.