Querying Counterfactuals on Tissue Graphs with Supervised Disentanglement
Abstract
Tissue graph counterfactuals ask what a cell would express if its spatial neighborhood were altered, a question central to predicting cell behavior in tissues. Existing methods, however, present only specific instances of such queries. In this work, we first formalize tissue graph counterfactuals as a class of spatial interventions that either rewire connections between cells (edge perturbation) or modify the expression of their neighbors (node perturbation). We then introduce Cellina (https://cellina.readthedocs.io), a framework that uses supervised disentanglement to decompose a cell's intrinsic state from its spatial context, using the latter as a conditioning input for in-silico perturbation predictions. Across benchmarks spanning over 2.5 million spatially-resolved cells in observational and interventional screens, Cellina outperforms both spatially-informed and non-spatial competitors in in-silico perturbations, while training as fast as non-spatial methods. Moreover, we show that Cellina reveals cancer subdomains with distinct signaling programs in an unsupervised manner and enables targeted neighbor perturbation simulations.
Related
- An Expert-Curated Benchmark for Frontier and Local LLMs Querying a Biomedical Knowledge Graph
- CPathOGen: Spatially and Morphologically Controlled H&E Counterfactuals for Probing Pathology Models
- scDisentangle: sequential disentanglement for single-cell perturbation response prediction
- Disentanglement of persistent and dynamic neural substrates of working memory
- Learning from Error Predictions Rather Than Prediction Errors: A Perspective on Self-Supervised Learning in the Brain
BibTeX
@misc{moeed2026querying,
title = {{Querying Counterfactuals on Tissue Graphs with Supervised Disentanglement}},
author = {Abdul Moeed and Stefan Schrod and Martin Rohbeck and Marc Jan Bonder and Pavlo Lutsik and Oliver Stegle and Daniel Dimitrov},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2606.08493},
eprint = {2606.08493},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2606.08493},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Querying Counterfactuals on Tissue Graphs with Supervised Disentanglement
AU - Abdul Moeed
AU - Stefan Schrod
AU - Martin Rohbeck
AU - Marc Jan Bonder
AU - Pavlo Lutsik
AU - Oliver Stegle
AU - Daniel Dimitrov
PY - 2026
DA - 2026/10/30
PB - arXiv
DO - 10.48550/arxiv.2606.08493
UR - https://arxiv.org/abs/2606.08493
AB - Tissue graph counterfactuals ask what a cell would express if its spatial neighborhood were altered, a question central to predicting cell behavior in tissues. Existing methods, however, present only specific instances of such queries. In this work, we first formalize tissue graph counterfactuals as a class of spatial interventions that either rewire connections between cells (edge perturbation) or modify the expression of their neighbors (node perturbation). We then introduce Cellina (https://cellina.readthedocs.io), a framework that uses supervised disentanglement to decompose a cell's intrinsic state from its spatial context, using the latter as a conditioning input for in-silico perturbation predictions. Across benchmarks spanning over 2.5 million spatially-resolved cells in observational and interventional screens, Cellina outperforms both spatially-informed and non-spatial competitors in in-silico perturbations, while training as fast as non-spatial methods. Moreover, we show that Cellina reveals cancer subdomains with distinct signaling programs in an unsupervised manner and enables targeted neighbor perturbation simulations.
ER -