Gene expression data holds the potential to shed light on multiple biological processes at once. However, data analysis methods for single cell sequencing mostly focus on finding cell clusters or the principal progression line of the data. Data analysis for spatial transcriptomics mostly addresses clustering and finding spatially variable genes. Existing data analysis methods are effective in finding the main data features, but they might miss less pronounced, albeit significant, processes, possibly involving a subset of the samples. In this work we present SPIRAL: Significant Process InfeRence ALgorithm. SPIRAL is based on Gaussian statistics to detect all statistically significant biological processes in single cell, bulk and spatial transcriptomics data. The algorithm outputs a list of structures, each defined by a set of genes working simultaneously in a specific population of cells. SPIRAL is unique in its flexibility: the structures are constructed by selecting subsets of genes and cells based on statistically significant and consistent differential expression. Every gene and every cell may be part of one structure, more or none. SPIRAL also provides several visual representations of structures and pathway enrichment information. We validated the statistical soundness of SPIRAL on synthetic datasets and applied it to single cell, spatial and bulk RNA-sequencing datasets. SPIRAL is available at https://spiral.technion.ac.il/ .