A Foldit plugin for running Apple's SimpleFold machine-learning model for protein structure prediction, as an in-game backend.
Given a sequence, SimpleFold predicts a 3D structure; this plugin runs that model and feeds the result back into Foldit as a molecular assembly the player can then fold and refine.
This is a kind = "python" plugin, hosted out-of-process through
foldit-python-host; the entry module is simplefold_plugin. The model code
itself lives in the ml-simplefold dependency under deps/, imported by the
thin plugin wrapper. The wrapper converts the model's output directly into molex
assembly wire bytes (see simplefold_plugin/molex_io.py), with no intermediate
structure-file format.
Model weights are not committed; they are resolved at load time.
Python plugins use pixi for their environment. Set it up through the workspace xtask:
cargo xtask setup-plugins simplefoldThe plugin subclasses the PluginInterface base class from the Foldit plugin
SDK; see the Foldit workspace docs ("Python and Native Plugins") for how the
host discovers and loads a Python plugin.