team 2: optimizing a simulator of molecular interaction for a GPU pipeline
To date, there are no standard methods to measure interaction rates between molecules in their native cellular environment. Rather, most interaction rates are measured in vitro. But how in vitro rates relate to rates in cells is unknown. Single-molecule light microscopy imaging provides a unique opportunity to capture individual molecular interaction events in their native cellular environment. However, single-molecule imaging is limited by sub-stoichiometric labeling, as a result of which only a small subset of interaction events is observed. We are developing a mathematical inference approach to extract the full population interaction rates from the subset of interaction events observed in single-molecule experiments. The approach consists of building a mathematical model of the biological system of interest and then running many simulations of the model while varying its parameters, including the unknown interaction rates, to match the experimental single-molecule data. Because the model is explicit in both space and time, these simulations are very time consuming. The goal of this project is to re-write and optimize the simulation code to run on a GPU instead of a CPU in order to speed up the simulations. Existing simulations with the original MatLab code can be used to verify the new simulations. Speeding up the simulations is going to be instrumental for moving this mathematical inference project forward.