Kandinsky: Enabling neighbourhood analysis of spatial omics data for functional insights on cell ecosystems

Spatially resolved omics technologies enable investigation of cellular interactions within their local microenvironments (neighbourhoods) directly in situ. Although several computational methods have been developed for neighbourhood analysis, significant limitations remain in how neighbourhoods are defined and interrogated. Here, we present Kandinsky, a toolkit that provides a flexible and versatile framework for defining and analysing cell neighbourhoods.

We developed Kandinsky to improve flexibility in neighbourhood analysis and maximise compatibility with a wide range of spatial omics data. We therefore implemented multiple approaches for identifying cell- or spot-based neighbourhoods that serve as input for four analytical modules: differential gene or protein expression analysis, neighbourhood clustering, co-localisation/dispersion, and spatial hot and cold areas. In addition to its core functionality, Kandinsky enables the execution of external tools within the same analytical framework.

We applied Kandinsky to real and simulated spatial datasets to benchmark its performance against existing methods and demonstrate its ability to uncover biologically meaningful spatial interactions. Kandinsky achieved competitive performance in terms of accuracy, memory usage, and runtime. In real datasets, it suggested transcriptional changes associated with acinar-to-beta cell reprogramming in the healthy pancreas; recapitulated stromal, immune, and tumour-associated clusters in pancreatic cancer; revealed the spatial co-localisation of myoepithelial cells with specific breast cancer subpopulations; and confirmed the association between regions of high CD74 expression and immune cell infiltration.

Kandinsky is a flexible and versatile toolkit for neighbourhood analysis that facilitates the exploration and interpretation of complex spatial omics data.