Fig. 5: Metabolic features improved the performances of clinical markers in entire and stratified participants.
From: Metabolic phenotyping reveals an emerging role of ammonia abnormality in Alzheimer’s disease
The auROC values (mean with S.E. from 7-fold cross validation) of gradient boosting models using basic (age, sex, BMI, APOE-ε4, and education year; blue) and combined (basic and 13 metabolic features; red) features for the differentiation of every 2 stages, in all (a) and stratified (b–g) participants. The sample numbers for NC, SCD, MCI, and AD are 487, 239, 284, and 387 respectively, for all (a) and stratified analysis (b:165, 69, 99, 154; c: 322, 170, 185, 233; d: 255, 122, 110, 90; e: 393, 197, 200, 203; f: 94, 42, 84, 184; g: 232, 117, 174, 297). h Scatter plot of whole brain Aβ deposition level and output of gradient boosting regression model with (red, Spearman correlation coefficient r = 0.47, p = 1.0E-15, two-sided) and without (blue, r = 0.21, p = 1.4E-5, two-sided) metabolic features (n = 421). The lines are linear fitting lines with 95% CI. Source data are provided as a Source Data file.
