Figure 2.

Naive and memory T cell proportions were the main differentiator between 4mo and 12mo mice. The importance of each immune factor in predicting host age was estimated by measuring the average gain in the purity of daughter nodes when decision tree splits used that feature for classifying all 32 mice mice as young or old irrespective of their infection status. Thus, the greater the importance of an immune feature, the more useful it is in classifying a mouse as 4mo or 12mo. We used a cut-off of 3 standard errors (red-shaded zone) above the mean (red dotted line) of all importances below which features were not considered robust enough for further inference. (a) Top predictors of age in mice, whether infected or not, achieved 97% accuracy on average over 10 repeated random training sets in correctly clarifying mice as 4mo or 12mo. Horizontal bars represent means and associated errors of the feature importances generated from 10 models trained on different nested random 75%/25% train/test repeats. Top predictors included the proportions of naive and memory T cells in the spleen as expected under thymic involution (see Supplementary Fig. S3): in the spleen, (b) the proportion of naive T helper cells (CD3+ CD4+ ) had decreased, whereas (c) the proportions of memory T cells exceeded 80% of all splenic T cells in 12mo mice while ranging between 55% and 72% in 4mo mice. (d) Among HSPC within the bone marrow, Lin⁻c-Kit⁺Sca-1⁻ (LK) cells were found in greater proportions in 12mo than in 4mo mice. See methods section for details on feature importance calculations. Horizontal bars represent means and associated errors of the feature importances generated from models trained on and testing against 10 repeated 75%/25% random splits of the full dataset. In boxplots, horizontal lines represent the group median, boxes the interquartile range, whiskers the overall range, and points represent outliers (<± 1.5 × the interquartile range, n = 16 per mice class).