Statistical Analysis of Tumor Phenotypes

Two different statistical tests were used to determine if the stratification procedures generated four different tumor phenotypes, as shown in Figure 2A and Figure 5A.
Firstly, the median enzyme/receptor transcript levels in the four ‘PG’ or ‘Bile’ groups were tested against the expected global population using chi-squared tests to answer the research question: Are the transcript medians of the four groups different from the population from which they were taken?
Secondly, the distribution of the enzyme/receptor transcript levels within each of the prostaglandin or bile synthesizing groups was compared with their distribution within the respective ‘Low PG’ or ‘Low Bile’ group using t-tests to answer the research question: Is the distribution of the transcript levels in the PG-positive or Bile-positive groups different from the respective ‘Low PG’ or ‘Low Bile’ groups?
Each patient tumor was identified as belonging to one of the three canonical GMP subtypes (classical, mesenchymal, or proneural) independently using the algorithms described on the Gliovis platform. Then, utilizing the contingency tables, chi-squared tests were performed to analyze the relationships between the distribution of these three canonical classifications within each of the four stratified ‘PG’ or ‘Bile’ groups and the global population distribution (Figure 2B and Figure 5B).
The distribution of cell-type proxies and individual gene transcripts were tested for statistical significance for each of the stratified prostaglandin or bile-synthesizing groups relative to their respective ‘Low PG’ or ‘Low Bile’ groups using t-tests to identify the statistical differences between normalized transcript levels in the phenotypes vs. the ‘Low PG’ or ‘Low Bile’ groups (Figure 3B, Figure 4, Figure 6B and Figure 7).
Kaplan–Meier survival curves were analyzed using the log-rank test, with statistical significance calculated from the chi-squared value compared to the ‘Low PG’ subgroups (Figure 4A) and the ‘Low Bile’ subgroups (Figure 6A).
In Tables S1–S5, the overlap between the stratified prostaglandin and bile tumors is compared with previously identified tumor subtypes; the tumors were stratified with respect to Treg infiltration and HIF markers or were based on tumor sex-steroid generation [1 (link)]. Chi-squared tests were also used to analyze the statistically significant overlap between the groups.
Excel was used to perform the chi-squared tests, the t-tests, and the Kaplan–Meier log-rank tests.
For the cross-correlation analysis, shown in Figures S1–S7 and S10, the Gliovis platform was used to calculate Pearson’s correlation between the transcript pairs as indicated by the numeral and the superscripted stars (*); statistical significance is indicated when p < 0.05 *, p < 0.01 **, and p < 0.001 ***.