Publications & Preprints¶

This work¶

Federated MaxFuse: Diagonal Integration of Weakly Linked Spatial and Single-cell Data through Federated Learning Krzysztof Baran, Swier Garst, Marcel Reinders, Jérémie Decouchant (2025) Delft University of Technology · MSc Computer Science thesis · 36 pp.

Integrating single-cell multi-omic data is crucial for comprehensive biological discovery, yet it remains challenging due to the weak correlation between modalities, data heterogeneity, and stringent privacy regulations. This thesis introduces Federated Matching xcross modalities via Fuzzy smoothed embeddings (MaxFuse), a novel adaptation of MaxFuse within a Federated Learning framework, which enables privacy-preserving diagonal integration through fuzzy smoothing, federated Canonical Correlation Analysis (CCA), and iterative matching without exchanging raw data.

The method this builds on¶

Integration of spatial and single-cell data across modalities with weakly linked features Shuxiao Chen, Bokai Zhu, Sijia Huang, John W. Hickey, Kevin Z. Lin, Michael Snyder, William J. Greenleaf, Garry P. Nolan, Nancy R. Zhang, Zongming Ma Nature Biotechnology 42, 1096–1106 (2024) DOI: 10.1038/s41587-023-01935-0 Open Access (CC BY 4.0)

Key results: 20–70% relative improvement under weak linkage; ~90% level-1 accuracy at a 30-antibody panel where competing methods reach 15–75%; the only method able to produce a structure-preserving joint embedding of CODEX and scRNA-seq data.

Datasets¶

Dataset	Source
CITE-seq PBMC	Hao et al. (2021), Cell 184:3573–3587 — Integrated analysis of multimodal single-cell data
CODEX tonsil	Kennedy-Darling et al. (2021), Eur. J. Immunol. 51:1262–1277
scRNA-seq tonsil	King et al. (2021), Sci. Immunol. 6:eabh3768

Both benchmark collections are distributed by the MaxFuse authors via the Wharton Research Data Services repository.

Methods referenced¶

Topic	Reference
Canonical correlation analysis	Hotelling (1936), Biometrika 28:321–377
Two-block Mode B PLS	Wegelin (2000), A survey of Partial Least Squares methods
PLS modelling	Wold (1975); Geladi (1988)
Leiden clustering	Traag et al. (2019), Sci. Rep. 9:5233
Jonker–Volgenant assignment	Crouse (2016), IEEE Trans. Aerosp. Electron. Syst. 52:1679–1696
HVG selection (Seurat V3)	Stuart et al. (2019), Cell 177:1888–1902
FOSCTTM	Liu et al. (2019), WABI
ASW-F1 / ARI-F1	Tran et al. (2020), Genome Biol. 21:12
Steadiness & Cohesiveness	Jeon et al. (2022), IEEE TVCG
ZADU evaluation library	Jeon et al. (2023)
CheckViz	Lespinats & Aupetit (2011), Comput. Graph. Forum
Kruskal's Stress	Kruskal (1964), Psychometrika 29:1–27
HDBSCAN	Campello et al. (2013), LNCS 7819

Citing¶

@mastersthesis{baran2025fedmaxfuse,
  title  = {Federated MaxFuse: Diagonal Integration of Weakly Linked Spatial
            and Single-cell Data through Federated Learning},
  author = {Baran, Krzysztof and Garst, Swier and Reinders, Marcel
            and Decouchant, J{\'e}r{\'e}mie},
  school = {Delft University of Technology},
  year   = {2025}
}

@article{chen2024maxfuse,
  title   = {Integration of spatial and single-cell data across modalities
             with weakly linked features},
  author  = {Chen, Shuxiao and Zhu, Bokai and Huang, Sijia and Hickey, John W.
             and Lin, Kevin Z. and Snyder, Michael and Greenleaf, William J.
             and Nolan, Garry P. and Zhang, Nancy R. and Ma, Zongming},
  journal = {Nature Biotechnology},
  volume  = {42}, pages = {1096--1106}, year = {2024},
  doi     = {10.1038/s41587-023-01935-0}
}