Yes. Pyxa 3D data can be analyzed without cell segmentation, which helps when cell boundaries are hard to define. We offer two complementary segmentation-free approaches:
1. LDA topic modeling (bin-based). The tissue is divided into small 20 µm bins, and a Latent Dirichlet Allocation model asks what cell-type expression signatures are present in each bin. Each bin is described as a mixture of signatures rather than forced into a single cell. In a mouse brain sagittal section, this revealed clear organization of the major brain regions, distinct neuronal and non-neuronal populations, and smooth transitions between regions that match known anatomy. It is well suited to spatial domain discovery, pathway activity mapping, and biomarker identification.
2. SM3DD (Spatial Molecular 3D Distance). This method works directly from the raw XYZ coordinates of each detected transcript. It computes pairwise 3D distances between molecules in continuous space and uses these distances, instead of expression values, as the PCA input. Short distances reveal high-density clusters, such as signaling hotspots and tissue niches. We have used it to recover density hotspots for ligand-receptor pairs, including opioid signaling, GABAergic co-localization, and dopamine signaling, without segmentation. The method is described in Blick et al., NAR Genomics and Bioinformatics 2026, 8, lqag007.
In short, LDA is best for tissue-level organization and cell-type signature mapping, and SM3DD is best for molecular niches and ligand-receptor co-localization.
Please note that neither approach gives single-cell resolution. If your question needs single-cell calls, segmentation-based analysis is recommended, and the approaches can be combined on the same dataset. We’re happy to discuss which fits your sample and goals.
