Extended Data Fig. 9: The prognosis association of cell types.
Related to Fig. 7. (a) Survival associations based on the signature genes of conventional and regulatory lymphocytes, B lymphocytes, and endothelial subsets per cell type. At the top, a bar plot elegantly displays the pan-cancer survival associations across 23 different cancer types, aggregated harmoniously using Stouffer’s method. Columns are gracefully ordered by combined Z-score. At the bottom, cancer-specific survival associations are tastefully presented, and statistical significance is calculated using the Cox Proportional-Hazards Model (b) Violin plots illustrating the signature scores of different lineages across five distinct tumor ecosystems, comprising a total of 8,734 samples from the TCGA project. The shape of the violin plot illustrates the distribution density of the data. The wider sections indicate a higher concentration of data points near that value (c) Box plots depicting the exhausted signature scores across three tumor ecotypes, comprising a total of 4,882 samples from the TCGA project. Statistical significance was evaluated using the two-tailed unpaired Wilcoxon tests combined with a permutation test (10,000 resamplings) to compare the signature score distribution across pairwise ecotypes; horizontal connectors denote compared groups, with the corresponding p-value indicated above the horizontal line. The bottom of the box represents the Q1, and the top of the box represents the Q3. The height of the box represents the IQR, while the horizontal line inside the box indicates the median. The whiskers extend to the positions of Q1 - 1.5 * IQR and Q3 + 1.5 * IQR. (d) Box plot depicting the CD8+ T cell infiltration among five patient subtypes. The CD8+ T cell infiltration was estimated using Cibersort from 8,734 samples from the TCGA project. Statistical significance of CD8+ T cell infiltration across pairwise patient subtypes was assessed using the two-tailed unpaired Wilcoxon tests. Horizontal lines connect compared groups, with corresponding p-values indicated above each line. The bottom of the box represents the Q1, while the top represents the Q3. The height of the box indicates the IQR, and the horizontal line inside the box represents the median. The whiskers extend to the positions of Q1 - 1.5 * IQR and Q3 + 1.5 * IQR. (e) Bar plot showing the proportion of each tumor sample ecosystem in different molecular cancer groups. (f) Heatmap displaying the average signature scores of fibroblast-associated cell states in each fibroblast cluster from the BRCA_GSE176078 dataset, with cell states identified separately by TabulaTIME (on the left) and Luca (on the right). (g) Left: Bar plot depicting the accuracy of using SCINA predicted annotations, based on distinct marker gene lists, versus annotation provided by the original published papers. Right: Bar plot depicting the normalized mutual information (NMI) of using SCINA predicted annotations, based on distinct marker gene lists, versus the cluster label. The datasets, listed from left to right, include the following number of samples and cells: 11 patients with 37,936 cells, 26 patients with 89,471 cells, 22 patients with 9,544 cells, 42 patients with 82,267 cells, and 12 patients with 90,603 cells. Significance was assessed using a two-tailed unpaired Wilcoxon tests and adjusted using the Benjamini-Hochberg (BH) method. The resulting p-values were 0.016 for ALL_GSE132509, 0.00058 for BRCA_GSE176078, 0.055 for ESCA_GSE173950, 0.026 for NSCLC_GSE148071 and 0.48 for RB_GSE166173. The bottom of the box represents the Q1, and the top of the box represents the Q3. The height of the box represents the IQR, while the horizontal line inside the box indicates the median. The whiskers extend to the positions of Q1 - 1.5 * IQR and Q3 + 1.5 * IQR (h) Left: Alluvial diagram showing changes in patient subtypes composition between TabulaTIME and Bagaev et al identified. Right: Scatter plot delineating the accuracy of different studies in distinguishing hot and cold tumors.
