Implementation of the Isokann algorithm. For a reference see our paper https://doi.org/10.1063/5.0140764 . Currently things are still fluctuating, so we have different implementations
- forced/isokann.jl - Isokann with adaptive sampling and optimal control for overdamped Langevin systems
- isomolly.jl - Isokann with adaptive sampling for Molly.jl systems (e.g. proteins)
- isosimple.jl - attempt at a cleaner version of isomolly.jl
- iso2.jl - Isokann 2 with multivariate memberships
After installing the package with Pkg.add("https://github.com/axsk/Isokann.jl") run a basic alanine dipeptide run with
using Isokann
iso = IsoRun() # create the Isokann system/configuration
run!(iso) # run it for iso.nd steps
plot_training(iso) # plot the training overview
# scatter plot of all initial points colored in corresponding chi value
scatter_ramachandran(iso.data[1], iso.model)
# estimate the eigenvalue, i.e. the metastability
eigenvalue(iso.model, iso.data[1])
The big todos:
- Merge optimal control into isomolly (requires a new Molly.jl sampler, almost done)
- Use PCCA+ for higher dimensional chi functions (almost done)
- Gather experience on larger molecules.
- refactor
The most recent (but not actually recent) plot:
