![[2507.04704] SPATIA: Multimodal Generation and Prediction of Spatial Cell Phenotypes](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2507.04704] SPATIA: Multimodal Generation and Prediction of Spatial Cell Phenotypes
Summary
The paper introduces SPATIA, a novel multimodal model for predicting spatial cell phenotypes by integrating cellular morphology, gene expression, and spatial context, achieving improved performance over existing methods.
Why It Matters
This research addresses a significant challenge in biology by providing a unified approach to analyze complex biological data. SPATIA enhances the understanding of tissue function and cellular behavior, which is crucial for advancements in biomedical research and therapies.
Key Takeaways
- SPATIA integrates multiple data modalities to predict spatial cell phenotypes.
- The model improves generative fidelity by 8% and predictive accuracy by up to 3% compared to existing methods.
- A multi-scale dataset of 25.9 million cell-gene pairs was utilized for benchmarking.
- The approach includes a novel spatially conditioned generative framework.
- SPATIA facilitates the modeling of microenvironment-dependent phenotypic transitions.
Quantitative Biology > Quantitative Methods arXiv:2507.04704 (q-bio) [Submitted on 7 Jul 2025 (v1), last revised 16 Feb 2026 (this version, v2)] Title:SPATIA: Multimodal Generation and Prediction of Spatial Cell Phenotypes Authors:Zhenglun Kong, Mufan Qiu, John Boesen, Xiang Lin, Sukwon Yun, Tianlong Chen, Manolis Kellis, Marinka Zitnik View a PDF of the paper titled SPATIA: Multimodal Generation and Prediction of Spatial Cell Phenotypes, by Zhenglun Kong and 7 other authors View PDF HTML (experimental) Abstract:Understanding how cellular morphology, gene expression, and spatial context jointly shape tissue function is a central challenge in biology. Image-based spatial transcriptomics technologies now provide high-resolution measurements of cell images and gene expression profiles, but existing methods typically analyze these modalities in isolation or at limited resolution. We address the problem by introducing SPATIA, a multi-level generative and predictive model that learns unified, spatially aware representations by fusing morphology, gene expression, and spatial context from the cell to the tissue level. SPATIA also incorporates a novel spatially conditioned generative framework for predicting cell morphologies under perturbations. Specifically, we propose a confidence-aware flow matching objective that reweights weak optimal-transport pairs based on uncertainty. We further apply morphology-profile alignment to encourage biologically meaningful image generation, enab...
