DATE:

Friday, June 11, 2021

TIME:

3:30 PM

LOCATION:

Virtual Zoom Conference

SPEAKER:

Eric Medwedeff, PhD Candidate, Computational Science, San Diego State University

ABSTRACT:

In our work, we aim to use machine learning to generate reduced models for complex biological systems described by dynamical graph grammars (DGG). So far, we have targeted the microtubule dynamics in the cytoskeleton of plant cells. In the cytoskeleton DGG, microtubules are represented as parameterized spatially embedded graphs with two flavors of local graph rewrite rules: (1) spatial parameters evolve in time subject to differential equations (2) structure changing events are determined stochastically. A simulation algorithm for these rule types can be derived using a DGG operator algebra based on a variation of the Chemical Master Equation (CME). For large simulations, however, a serial version quickly becomes intractable. A parallel version is more tractable, but also very costly. Thus, reinforcing the need for a reduced model. Currently, we are in early stages of developing a parallel DGG simulation algorithm. We have a prototype known as Cajete written using the open source national lab molecular dynamics library Cabana and the performance portable back-end library, Kokkos. In the future, our goal is to use the simulation output from Cajete as means to obtain massive amounts of training data to generate a reduced model for the cytoskeleton dynamics. We then plan to extend this framework and use it as a template for generating reduced models for a wider class of problems.

HOST:

Abraham Flores and the SDSU SIAM Student Chapter

VIDEO: