Journal: Research
Article Title: Deep Learning-Enabled Integration of Histology and Transcriptomics for Tissue Spatial Profile Analysis
doi: 10.34133/research.0568
Figure Lengend Snippet: GIST enhanced the identification of marker genes from differentially expressed genes (DEGs) in human breast cancer. (A) Spatial regions predicted by GIST (ARI = 0.61), CellCharter (ARI = 0.40), and PROST (ARI = 0.21) compared to ground truth. GIST predictions exhibit strong alignment with the ground truth, particularly in accurately delineating tumor regions. (B) In the raw UMAP, tumor cells in red dots are scattered, making them difficult to distinguish from other cell types. After enhancement, tumor cells are clustered together, improving their identification and separation. (C) Spatial visualization of raw and enhanced ERBB2 expression in an FFPE human breast sample. The enhanced visualization reduces noise and reveals clearer patterns of ERBB2 expression ( t test, P = 0.00084), demonstrating the effectiveness of the enhancement method. (D) GIST-enhanced gene expression patterns improve the identification of marker genes associated with specific biological processes or cancer states, such as ERBB2. (E) Visualization and comparison of raw (purple) and enhanced (orange) ESR1 expression across different cell clusters highlight expression patterns and the number of DEGs that are not apparent in the raw data but become evident with enhancement.
Article Snippet: This dataset was derived from 10x Genomics on FFPE human breast tissue sourced from BioIVT’s Asterand Human Tissue Repository.
Techniques: Marker, Expressing, Gene Expression, Comparison