Fig. 4

Integrated prediction model combining microbial and immunological signatures stratified according to response to nCCRT. a Venn plot of candidate microbes for prediction model construction. Candidate microbes were derived from 16 genera that differed between the good- and poor-response groups identified in cross-sectional analysis (LEfSe analysis before and during nCCRT), 11 different genera sourced from longitudinal analysis (ZIBR analysis), and 136 genera that contributed to microbial clustering in multivariate statistical analysis (sPLS before and during nCCRT). Finally, 6 genera (cyan background) were used for model construction. b Hierarchical clustering plot of correlation between baseline levels of candidate genera and log2-transformed fold-change of immune cells and immunomodulatory proteins from before to during nCCRT. Color scale represents the Spearman rank correlation coefficient. c Areas under the receiver operating characteristic (ROC) curve (left) and the precision-recall curve (right) to measure model performance in the validation set, with variables and coefficients as shown. d Evaluation of differences in log2-transformed fold-change of Intestinimonas from before nCCRT to during nCCRT and the baseline level of Clostridium sensu stricto 1 between the good- and poor-response groups. e. Scatter plot of baseline level of Intestinimonas showed positive association with log2-transformed fold-change in PD-1 levels from before to during nCCRT. *P < 0.05; **P < 0.01; ***P < 0.001