Installation

Julia

Install Julia (v1.11 or newer) with juliaup:

curl -fsSL https://install.julialang.org | sh

After restarting your shell, julia starts the REPL.

ISOKANN

In the REPL, enter package mode (type ]) and add the release version:

pkg> add ISOKANN
pkg> test ISOKANN

For the latest development version:

pkg> add https://github.com/axsk/ISOKANN.jl

If you are new to Julia, take a moment to read about Pkg environments; ISOKANN is most convenient when used inside a project environment.

The first run precompiles a sizeable dependency tree — expect a few minutes.

OpenMM (for molecular dynamics)

Since v2.0, ISOKANN calls OpenMM through PythonCall.jl, and Python dependencies are managed by CondaPkg.jl. Installing ISOKANN should bring OpenMM, joblib, and openmmforcefields in automatically — no manual PyCall/Conda build step is needed.

If you want the same conda environment available from a Python shell too (e.g. to use the Python bindings in python/), point your tooling at the environment CondaPkg provisions (typically under .CondaPkg/env/), or share a single conda environment via CondaPkg's CONDAPKG_BACKEND / shared-env settings.

For troubleshooting, see the CondaPkg and PythonCall docs, or open an issue on the ISOKANN repository.

Development

To hack on ISOKANN itself:

git clone git@github.com:axsk/ISOKANN.jl.git
cd ISOKANN.jl
julia --project=.

Then inside Julia:

pkg> instantiate

We strongly recommend loading Revise before ISOKANN so edits take effect without restarting:

using Revise
using ISOKANN

See the Julia Pkg docs for a deeper tour of environments and development workflows.

Editor

VS Code with the Julia extension works well. On a remote machine, the Remote-SSH / Remote Coder extensions let you drive the same setup locally. The package is designed around interactive use in the REPL, so keep one open alongside your editor.

GPU

Training and most analysis routines run on CUDA GPUs via Flux/Functors. Move an Iso to and from GPU with gpu(iso) / cpu(iso). The OpenMM simulation itself selects its own platform (CPU/CUDA/OpenCL) via OpenMM's platform mechanism.