Physics-based machine learning for stochastic reaction networks

Methods and papers related to physics-based maching learning for modeling stochastic reaction networks, as developed in the thesis of Oliver K. Ernst

Research groups:

Eric Mjolsness Departments of Computer Science and Mathematics, UC Irvine

Terrence Sejnowski & Tom Bartol Computational Neurobiology Lab, Salk Institute for Biological Studies

GitHub:

Click here to browse all code through this organization's GitHub page.

Thesis

Modeling reaction-diffusion systems with dynamic Boltzmann distributions

Papers & Code

Physics-based machine learning for modeling stochastic IP3-dependent calcium dynamics

Deep learning moment closure approximations using dynamic Boltzmann distributions

Learning moment closure in reaction-diffusion systems with spatial dynamic Boltzmann distributions

Learning dynamic Boltzmann distributions as reduced models of spatial chemical kinetics

Graph-constrained Correlation Dynamics (GCCD)
- Paper: Model reduction for stochastic CaMKII reaction kinetics in synapses by graph-constrained correlation dynamics
- Code: Graph-constrained Correlation Dynamics (GCCD)

Software

Various codes attached to papers (see above).

TensorFlow Python package for physics-based dynamic PCA

C++ library for Dynamic Boltzmann Machines for Lattice Chemical Kinetics

C++ library for Gillespie simulations on a lattice

C++ library for Gillespie simulations

Python package to sample pairs of particles from discrete Gaussians