Fig. 5: Spatial biomarkers.
From: A deep learning-based multiscale integration of spatial omics with tumor morphology
a Example slide from a TCGA patient with basal breast cancer and poor predicted prognosis. Scale bar: 2 mm. b Local survival scores output by the model trained to predict overall survival. Shades of blue indicate areas associated with a lower risk by the model, while red indicates areas associated with a high risk. c Predicted expression of IGKC, MKI67 and SCUBE2 on the same slide, normalized between 0 and 1. d Bar plot showing the DR scores of 8 example genes in good prognosis patients with different molecular subtypes from TCGA-BRCA; n(luminal A) = 143, n(luminal B) = 60, n(HER2) = 20, n(basal) = 46. Boxes represent the average value over slides, while error bars indicate the 95% confidence interval obtained by bootstrapping 10,000 times over samples. e Same as d, with intermediate prognosis patients; n(luminal A) = 284, n(luminal B) = 120, n(HER2) = 40, n(basal) = 92. f Same as d. and e., with poor prognosis patients; n(luminal A) = 143, n(luminal B) = 61, n(HER2) = 20, n(basal) = 47. Source data are provided as a Source Data file.
