Spatial transcriptomics (ST) is an exciting area of research that lets scientists study which genes are active in different parts of a tissue. Imagine being able to zoom into tiny sections of tissue, almost like looking at a map, and seeing how cells behave based on their gene activity. This can give us a lot of information about how diseases develop and how different areas of tissue function. However, studying this information at a very high resolution, especially when looking at large, complex tissues like muscles or blood vessels, can be really challenging.
One of the main problems is that traditional methods rely on “cell segmentation,” which is a fancy way of saying they try to outline the borders of each cell. This works well when cells are neatly shaped and evenly sized, but in real tissues, cells can be irregular and messy, making it difficult for existing methods to get an accurate picture of what’s going on at the gene level. On top of that, when analyzing large-scale tissue samples, you’re dealing with billions of data points, which can quickly become overwhelming for current methods.
FICTURE, is a new method developed by researchers at the University of Michigan and introduced in this study. Unlike traditional methods, FICTURE doesn’t need to outline individual cells (segmentation-free). Instead, it analyzes the gene activity directly from the tissue as a whole and breaks it down into different “spatial factors,” or patterns of gene activity, across tiny pixels of tissue. Think of it as a super-efficient way to understand how genes are behaving across the tissue without needing to worry about cell borders.
Overview of FICTURE
a, Illustration of the ST analysis on a two-dimensional (2D) section of three-dimensional (3D) tissue. The 2D view of each cell varies substantially depending on the shape of the cells and the location and orientation of the cutting plane. A 2D slice may not capture some nuclei, leading to bias in nuclear segmentation based on histological staining (bottom left, dashed lines). Uniform gridding substantially compromises the original resolution (bottom center). FICTURE preserves the original resolution at the pixel level (bottom right). b, The proportion of transcripts included (yellow) or excluded (navy) in the cell segmentation analysis across five high-resolution datasets used in this study. Asterisk indicates that the Seq-Scope dataset was not segmented into cells. c, Schematic of the FICTURE’s workflow. FICTURE’s pixel-level inference is based on factor-specific expression profiles. These can be generated from (i) unsupervised factorization using LDA on spot-level gene counts collapsed according to a hexagonal grid (default; D, left), (ii) other tools that provide clustering or factorization (for example, Seurat, Scanpy and Squidpy) on collapsed data, or (iii) external single-cell/single-nucleus RNA-seq reference with cell-type-specific gene expressions. d, Schematic illustration of the FICTURE’s algorithm. Based on the initial factors, FICTURE places anchors on a lattice denser than cells (center) and infers a mixture distribution over factors at each anchor based on the gene expressed at pixels in its neighborhood. Each pixel is assigned to a nearby anchor probabilistically, and the pixel’s sparse gene expression is modeled by that anchor’s mixture distribution. Initially pixels are assigned to the nearest anchors deterministically, then the anchors’ mixture distributions over factors and the pixel-to-anchor assignments are updated iteratively. Upon convergence, pixels are assigned to anchors with factor mixtures best explaining the pixels’ gene expression; and anchors tend to collect information from a more homogeneous set of pixels (right).
FICTURE uses advanced mathematical modeling (specifically, something called a Dirichlet model) to handle these large datasets and is much faster than existing methods. The researchers behind FICTURE tested it on real tissue samples, including difficult-to-study areas like blood vessels and fibrotic tissue (where there’s a lot of scarring). The results showed that FICTURE could uncover gene expression details that other methods missed, making it a valuable tool for understanding complex tissue structures.
In short, FICTURE allows scientists to explore the tiny, detailed world of gene activity in tissues with unmatched precision and speed, helping advance research in everything from heart disease to muscle disorders. This method can work across different platforms and types of data, making it versatile and widely applicable in the growing field of spatial transcriptomics.
Availability – The source code and Python package for FICTURE method are publicly available in the GitHub repository at https://github.com/seqscope/ficture/.