Extended Data Fig. 9. 16S rRNA Random Forest Model Validation.
a) Total body scores (TBS) used to train a random forest model for prediction of PMI (ADD) shows that TBS scores can predict PMI relatively accurately based on a low MAE but have higher variability in their predictions as represented by a higher residual value than microbiome-based models. Models built from 16S rRNA data using SILVA level-7 taxa from the skin and soil associated with the hip were validated with b) an independent test set of samples that were collected from cadavers at locations and climates not represented in our model and c) the same data where samples were given randomly assigned ADDs within the range of true ADDs to serve as a null model. Significance measured with linear mixed-effects models within each location and adding a random intercept for cadavers with two-tailed ANOVA and no multiple comparison adjustments. Data are presented as mean values +/− 95% CI.