Statistical methods to delineate the spatial and temporal pattern of cell-cell interactions in spatial transcriptomics data

The National Institute of General Medical Sciences awarded $899,250 to The University of Texas MD Anderson Cancer Center on September 15, 2025, under the Biomedical Research and Research Training program (CFDA 93.859) to develop statistical methods that delineate spatial and temporal patterns of cell-cell interactions in spatial transcriptomics data.

The project addresses methodological gaps in assessing cell-cell interactions and their heterogeneity across spatial transcriptomics platforms. The recipient will develop a novel statistical model to estimate spatial cell-cell interaction patterns while accounting for spot-level cellular compositions and spatial location information, then extend this model to population scale to identify interaction patterns with differential heterogeneity across subjects with varying disease conditions. The work incorporates spatial and temporal constraints to detect differential interaction patterns in longitudinally collected spatial transcriptomics data, modeling subject-specific patterns and leveraging data from multiple time points. The recipient will generalize the statistical method by integrating serially sectioned spatial transcriptomics data with histological images to provide a robust solution for three-dimensional cell-cell interaction analysis.

The proposed methods will be applied to spatial transcriptomics data from pancreatic and gastric cancer patients at MD Anderson Cancer Center, as well as brain samples from patients with early- to late-stage Alzheimer's disease.

Performance takes place in Houston, Texas. The award period runs through July 31, 2030.

Abstract

Although studies of cell-cell interactions (ccis) have attracted substantial attention from researchers to understand cellular development and related disease onset, considerable methodological gaps remain in assessing ccis and their intra-sample heterogeneity in spatial transcriptomics (ST) data. This proposal is driven by our collaborative work with biologists and physicians to investigate spatially heterogeneous ccis and their implications across various human diseases. The primary objective of this proposal is to develop robust statistical methods to quantify the heterogeneous CCI patterns on ST platforms and identify spatial or temporal differentially behaved ccis across conditions. The project is structured around three complementary research directions. 1) we propose a novel statistical model for ST data to estimate spatial CCI patterns, overcoming limitations in existing methods. This model specifically accounts for spot-level cellular compositions and spatial location information. We will extend this model to a population scale to identify CCI patterns with differential heterogeneity across subjects with varying disease conditions. 2) we aim to incorporate spatial and temporal constraints into statistical models to detect differential CCI patterns in longitudinally collected ST data. This optimized approach will improve CCI quantification by modeling subject-specific patterns and leveraging ST data from multiple time points. 3) we will generalize our statistical method by integrating serially sectioned ST data with histological images, providing a robust solution for 3D CCI analysis. The proposed methods will be applied to ST data from pancreatic and gastric cancer patients at the university of texas MD anderson cancer center, as well as brain samples from patients with early- to late-stage alzheimer's disease. While initially motivated by studies of cancer and alzheimer's disease, these statistical methods are broadly applicable for estimating intra-sample heterogeneity of ccis in other disease contexts and in embryo development research. Once validated, we will make all software packages developed in this project available to the wider research community. The proposed methods will bring unprecedented analytical power to characterization of cellular interactions in their surrounding environments, allowing for discovery of many hidden phenotypes that are disease- or patient outcome-associated.