This project
MaxFuse
Two versions of upstream MaxFuse exist
The released PyPI package and the development snapshot archived in the repository
(Archive/MaxFuse_devo/) differ algorithmically, notably in the nearest-neighbour metric used
during propagation, while sharing a version string. If you benchmark against "MaxFuse",
record the exact commit or package version. See
Comparison with Centralized MaxFuse.
Upstream licensing
The reference implementation is distributed under the Stanford/Penn Academic Software
License: academic non-commercial use only, no redistribution without written permission,
plus a broad reverse-engineering and derivative-ownership clause. This is not an open-source
licence, and its scope is the reason Fed-MaxFuse's own licensing status is currently
unresolved. Check it before building on upstream code.
Core dependencies
Comparable integration methods
| Method |
Approach |
Notes |
| Seurat v3 |
Horizontal, anchor-based |
Strong with shared features; degrades under weak linkage |
| Harmony |
Batch correction |
Different problem; used before MaxFuse in multi-donor settings |
| LIGER |
iNMF |
Requires shared feature space |
| BindSC |
Bi-order integration |
Benchmarked in the MaxFuse paper |
| scGLUE |
Deep generative |
Comparable on strong linkage; far higher compute cost |
Federated learning
| Resource |
Relevance |
| Flower |
General FL framework — a candidate for a networked deployment |
| NVIDIA FLARE |
FL for healthcare |
| Kairouz et al. (2021), Advances and Open Problems in Federated Learning |
Survey; useful on threat models |
Infrastructure