Fig. 4: DSS temporal signature and trajectory predictive power of colonic histopathology across tissue, location, and colitis model.

The training data comprise DSS whole-colon temporal expression/splicing signatures and trajectories and V(D)J clonal deconvolution. The signatures are dimensionally reduced to the first three principal components and median expression score. A random forest model is trained on all of the DSS WC Janssen samples. Validation of the trained model was performed over different tissues, DSS experimental locations, and adoptive transfer colitis models to predict colonic histological scoring (or disease status). Prediction efficacy was evaluated via Spearman correlation value (and p value) and a corresponding Kappa value computed for disease status classification. DSS temporal signature and trajectory predictive power of colonic histopathology across tissue, location, and colitis model. a Heatmap of spearman correlation values (with value as color, significance as opacity and size) between predicted and observed histopathology. The relative predictor importance to each predictive model is indicated by the bar-charts. b Distribution of histopathology scores across models. Over 36 days the DSS models induce higher disease severity than either T-cell model.