Querying Counterfactuals on Tissue Graphs with Supervised Disentanglement

2026/06/07 by Abdul Moeed, Stefan Schrod, Martin Rohbeck +4 · 1 voice
Biochemistry, Genetics and Molecular Biology · Computer Science · Mathematics · #q-bio.GN #cs.LG #stat.ML

paper · pdf

arxiv published 2026/06/07 · arxiv updated 2026/06/10

Abstract

Tissue graph counterfactuals ask how a cell's expression would change under altered spatial neighbor contexts. Such queries are central to predicting cell behavior in tissues, but lack a unified definition, with existing methods targeting specific intervention types or treating cells as i.i.d. 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 counterfactual predictions. Across benchmarks spanning over 2.5 million spatially-resolved cells in colorectal cancer and mouse brain, Cellina outperforms spatially-informed and non-spatial competitors in in-silico graph perturbations, disentanglement, and scalability. Additionally, we show that Cellina reveals biologically distinct cancer subdomains in an unsupervised manner and enables targeted neighbor perturbation simulations.

Citations

• STRAND: Sequence-Conditioned Transport for Single-Cell Perturbations
• Interpretation, extrapolation and perturbation of single cells
• Integrated in vivo combinatorial functional genomics and spatial transcriptomics of tumours to decode genotype-to-phenotype relationships
• How to Build the Virtual Cell with Artificial Intelligence: Priorities and Opportunities
• PerturBench: Benchmarking Machine Learning Models for Cellular Perturbation Analysis
• Toward a foundation model of causal cell and tissue biology with a Perturbation Cell and Tissue Atlas
• Learning Causal Representations of Single Cells via Sparse Mechanism Shift Modeling
• CLEAR: Generative Counterfactual Explanations on Graphs
• Predicting Cellular Responses to Novel Drug Perturbations at a Single-Cell Resolution
• Disentanglement via Mechanism Sparsity Regularization: A New Principle for Nonlinear ICA
• Robust Counterfactual Explanations on Graph Neural Networks
• How Attentive are Graph Attention Networks?
• Toward Causal Representation Learning
• CF-GNNExplainer: Counterfactual Explanations for Graph Neural Networks
• Deciphering cell–cell interactions and communication from gene expression
• Benchmarking atlas-level data integration in single-cell genomics
• Supervised Contrastive Learning
• Variational Autoencoders and Nonlinear ICA: A Unifying Framework
• Demystifying Inter-Class Disentanglement
• Fast Graph Representation Learning with PyTorch Geometric
• Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations
• Variational Graph Auto-Encoders

Discussions

• Abdul Moeed, Stefan Schrod, Martin Rohbeck, Marc Jan Bonder, Pavlo Lutsik, Oliver Stegle, Daniel Dimitrov: Querying Counterfactuals on Tissue Graphs with Supervised Disentanglement https://arxiv.org/a [bsky, 0 points, 0 comments]

Related

• Celcomen: spatial causal disentanglement for single-cell and tissue perturbation modeling
• Context-dependent spatial multicellular network motifs for single-cell spatial biology
• SemanticST: Spatially Informed Semantic Graph Learning for Clustering, Integration, and Scalable Analysis of Spatial Transcriptomics
• TopROI: A topology-informed network approach for tissue partitioning
• Novel Methods for Analyzing Cellular Interactions in Deep Learning-Based Image Cytometry: Spatial Interaction Potential and Co-Localization Index