bioRxiv Science⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.1101/2025.01.25.634867

Building Foundation Models to Characterize Cellular Interactions via Geometric Self-Supervised Learning on Spatial Genomics

You, Y.·Wang, Z.·Fleisher, K.·Liu, R.·Thomson, M.
View original source ↗Export metadata ↓

Abstract

Cellular interactions form the fundamental/core circuits that drive development, physiology, and disease within tissues. Advances in spatial genomics (SG) and artificial intelligence (AI) offer unprecedented opportunities to computationally analyze and predict the behavior of cell intricate networks, and to identify interactions that drive disease states. However, challenges arise in both methodology and scalability: (i) how to computationally characterize complicated cellular interactions of multi-scale nature where chemical genes/circuits in individual cells process information and drive interactions among large numbers of diverse cell types, and (ii) how to scale up the pipeline to accommodate the increasing volumes of SG data that map transcriptome-scale gene expression and spatial proximity across millions of cells. In this paper, we introduce the Cellular Interaction Foundation Model (CI-FM), an AI foundation model functioning to analyze and simulate cellular interactions within living tissues. In the CI-FM pipeline, we explicitly capture and embed cellular interactions within microenvironments by leveraging the powerful and scalable geometric graph neural network model, and optimize the characterization of cellular interactions with a novel self-supervised learning objective - we train it to infer gene expressions of cells based upon their interacting microenvironment. As a result, we construct CI-FM with 100 million parameters by consuming SG data of 23 million cells. Our benchmarking experiments show CI-FM effectively infers gene expressions conditional on the microenvironmental contexts: we achieve a high correlation and a low mismatch error (MSE of 1.1% relative to the square median expression), with 79.4% of cells on average being annotated as the similar cell type based on their predicted and actual expressions. We demonstrate the downstream utility of CI-FM by: (i) applying CI-FM to embed tumor samples to capture cellular interactions within tumor microenvironments (ROC-AUC score of 0.76 on classifying sample conditions via linear probing on embeddings), and identifying shared signatures across samples; and (ii) using CI-FM to simulate changes in microenvironmental composition in response to T cell infiltration, which highlights how CI-FM can be leveraged to model cellular responses to tissue perturbations - an essential step toward constructing "AI virtual tissues". Our model is open source and publicly accessible at https://huggingface.co/ynyou/CIFM.

Explore related subjects

bioinformaticsPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

You, Y., Wang, Z., Fleisher, K., Liu, R., Thomson, M.. 2025-01-27. Building Foundation Models to Characterize Cellular Interactions via Geometric Self-Supervised Learning on Spatial Genomics. https://doi.org/10.1101/2025.01.25.634867

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

deepTools4: Faster, Scalable, and Modernized Genomic Data Analysis

The deepTools suite is widely used for quality control, summarization, and visualization of high-throughput sequencing data. However, as genomic datasets grow in scale and depth, existing tools often become computational bottlenecks. We present deepTools4, a modernized framework designed for high performance, scalability, and enhanced usability, while maintaining seamless integration with the Galaxy platform. Core functionality has been rewritten and thoroughly benchmarked, achieving substantial computational speed-ups. Visualization tools have been modernized to provide more intuitive, high-quality graphical outputs. In addition, our testing framework has been expanded to facilitate continuous development and ensure robustness. We demonstrate the utility of deepTools4 across diverse epigenomics and transcriptomics datasets, highlighting its ability to summarize genome-wide signals, apply a wide range of common normalization methods, and visualize enrichment patterns. The enhanced deepTools4 framework delivers a faster, more scalable solution for reproducible genome-wide analysis and interpretation across many sequencing modalities. Availability: https://github.com/deeptools/deepTools/

bioinformatics↗
bioRxiv2026

Reproducibility and cross-dataset concordance of drug-induced transcriptional responses

Drug-perturbation transcriptomic resources are expanding, but the reproducibility and cross-source consistency of the responses they measure remain critical to their use. We compared six resources using independent replicates, matched-drug analyses, and response decomposition. High agreement in treated expression did not translate into equally repeatable drug-induced changes. Cross-source response concordance was generally limited, including under matched cell and treatment conditions, although some comparisons retained drug-specific information. In Tahoe, a control-expression-only baseline left substantial drug-specific variation unresolved. Measured drug main effects - the average responses across cell lines - provided information for predicting cell-specific interactions, with this relationship retained in a cross-plate validation. Exploratory analyses of structure, target, and external-response representations identified modest signals in selected pipelines. Together, these analyses characterize the reliability and transferability of current perturbation resources and identify prediction of drug main effects as a key bottleneck in the framework examined, motivating shared reference measurements and better recovery of drug-specific responses.

bioinformatics↗
bioRxiv2026

Propensity-Adjusted AUC under Nuisance-Model Misspecification

Propensity adjustment compares medical-image detector scores within groups with similar measured properties, such as stain or focus. We examine its sensitivity to the nuisance model that forms those groups. Under Gaussian latent-index assumptions and exact linear-index strata, a closed-form correlation gives an exact boundary for adjusted AUC to fall below one half. Quadrature matches 36 simulation cells within 0.008. Inflation and attenuation relative to correct adjustment are empirical findings, not general ordering results. An overlap screen and a self-stratified control introduced for this study miss 140 of 141 simulated cells with absolute error above 0.05: checking the fitted propensity cannot certify completeness of the adjustment set. On real pathology corpora, logistic and boosted-tree models using the same 16 features disagree. Among splits where both pass both checks, mean adjusted-AUC gaps are 0.057 on acid-fast bacilli (AFB), 0.048 on Kaggle, and 0.049 on PCam, over three, five, and three splits respectively. These gaps measure model sensitivity; neither estimate is known to be correct. A controlled colour shift moves adjusted AUC relative to model-specific references in the predicted direction in 78 of 80 label-dependent cells. Broader balance and label-fit diagnostics favour trees without certifying their adjusted AUC. We recommend reporting sensitivity across nuisance models alongside these diagnostics. Reversal relative to known truth occurs in constructed stress tests; its frequency in practice remains unknown.

bioinformatics↗