Attempts at computational analysis of cell coordination at different scales

Coordinated cellular activity is fundamental to tissue function and is commonly perturbed in disease, yet scalable methods to quantify such coordination and its mechanisms remain limited. I will discuss these challenges and how my group has approached it at different scales: from analysis of population-scale transcriptional analysis of patient cohorts, down to local tissue contexts resolved by spatial transcriptomics. First, I will outline an interpretable tensor‑decomposition framework that we designed for analysis of population‑scale scRNA‑seq that captures multicellular expression programs - correlated state changes across cell types - and relates them to clinical or environmental covariates in patient cohorts. We can use such multicellular programs to look for factors potentially mediating this coordination and use genetic variation within population to evaluate causal hypothesis. Though the mechanisms underlying such organism-wide coordination are also likely to be indirect. In contrast, at the scale of small cellular neighborhoods within tissues, we expect more direct influences to be key. I will discuss how spatial transcriptomics can quantify coordination in situ and the challenges faced by such approaches. In particular, I will show that most current techniques are vulnerable to pervasive segmentation errors that misattribute transcripts across adjacent cells, generating spurious region‑specific signals, interaction‑changed genes, and ligand–receptor pairs. I will describe a practical mitigation strategy based on factorization of subcellular neighborhood composition vectors and probabilistic labeling to identify and remove admixture‑driven factors, thereby restoring biologically coherent signals and improving interpretability of downstream analyses. The seminar will thus connect global multicellular programs to spatially resolved mechanisms while delineating necessary controls for confounding.