Spatial Transcriptomic Tissue-Domain Identification

A 10x Visium slide measures gene expression in thousands of spots on a tissue section next to an H&E image. This project splits the spots of a section into tissue domains - the six layers of the human cortex and white matter - with graph neural networks (GCN and GAT autoencoders on the spot neighbour graph) and an H&E patch CNN, and compares them honestly with PCA clustering and simple spatial smoothing on 12 public spatialLIBD sections of 3 donors, tuned on two donors and tested on the third. Every spatial method beats expression-only clustering on all 12 sections (test ARI 0.28-0.30 against 0.19), but the GNNs only tie with simple smoothing, and the image separates white matter, not layers.

Category: Bioinformatics - Transcriptomics and Gene-Expression Analysis

Transcriptomics & Gene-Expression Analysis Project B16 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.