Fig. 4: GLASS-AI analysis uncovers intratumor heterogeneity.

a Representative annotations of tumor grades produced by human raters (left) and GLASS-AI (right). Scale bars represent 200 µm. b–d The intratumor heterogeneity of individual tumors in the K (n = 11) and TK (n = 13) mice was analyzed by examining the proportion of each tumor grade within a tumor. Each bar represents a single tumor with the proportion of each grade of LUAD within that tumor shown by the various colors. The black lines overlayed on the graphs and corresponding brackets under the graphs indicate the overall tumor grade assigned to each tumor based on the highest tumor grade present that comprises ≥10% of the tumor’s total area (b). The average Shannon Diversity Index (SDI) of each animal’s tumors was calculated and compared by a two-tailed Student’s t-test. Boxplots are presented in Tukey style; a line at the median with IQR, crosses indicating the mean, whiskers showing the lesser of 1.5× IQR or most extreme values, and individual values represented by points (c). Individual tumors were grouped by genotype and grade to perform a more in-depth analysis of the distribution of intratumor heterogeneity in these mouse models and analyzed by Kruskal–Wallis non-parametric test with two-tailed Dunn’s posthoc test due to high skewness and kurtosis of the SDI values (d). Lines indicate median values, and whiskers show IQR (d). *p < 0.05 for the indicated comparison between K and TK mice (c, d). Numerals indicate *p < 0.05 between the noted grade within the same genotype (d).