Persistent homology features (PHFs) for machine learning on chemistry and biochemistry problems. PHFs are compact topology based encodings of 3-D structure. Atoms are taken to be points in a point cloud and then the persistence diagrams are created from them and vectorised to a highly compact 1-D input. Using PHFs can solves molecular machine learning problems with results equal to the state of the art in a 10th or a 100th of the time, disc size and energy usage.
For a demo of the code and overview of the mathematics see: https://www.youtube.com/watch?v=7AMfpigTuVY
See the manual and the paper: E. Gale, 'Shape is (almost) all!: Persistent homology features (PHFs) are an information rich input for efficient molecular machine learning.' Forthcoming 2026
Note that this repo is currently in alpha and being updated. The paper will be uploaded to arXiv in the next month (April 2026)