Fig. 3: Later life outcome stratification.

a, Normalized stability plot, showing that the best clustering solution is k = 3. This solution achieves 99% accuracy when generalizing to the validation set. Error bars in this plot represent the 95% confidence interval for normalized stability from repeated cross-validation. b–d, UMAP scatter boxplots (c,d) and density plots (b) showing the high and low subtypes. The boxplots indicate IQR (first quartile, 25th percentile; third quartile, 75th percentile), and the whiskers indicate Q1 − (1.5 × IQR) or Q3 + (1.5 × IQR). The line within the boxplot represents the median. The plots for NDA Training (c) and NDA Validation (d) describe the subtypes for each of the three VABS domains (communication, living skills and socialization). e, Alluvial plot of the n = 130 subset of individuals that were in both early snapshot (left) and later life outcome (right) cohorts. The sample sizes for a–d were n = 1,715 for the NDA Training set and n = 846 for the NDA Validation set.