Fig. 3

Predicting future A\(\beta\)-positivity from multimodal data excluding PET and CSF baseline measures: a Bar plots showing the average AUC and average correlation score across 10 computation runs for CSF-based and PET-based models, with 95% confidence intervals error bars. b Distribution of probability score derived by RLR for PET-based and CSF-based prediction in A\(\beta\)-Stable and A\(\beta\)-Converter groups. c, d Scatter plot for estimation of A\(\beta\)42 (c) and global SUVR (d) derived by ridge linear regression (with demographics, APOE4, neuropsychology, and MRI biomarkers). The results in b, c, and d are from 1 computation run with median performance. e, f Heatmap of coefficient values across 10 runs of 10-fold CV (100 models) for CSF-based models (e) and pet-based models (f), with a single column heatmap representing the correlation score between each variable and the label (A\(\beta\)-Stable, A\(\beta\)-Converter), and a bar graph showing the importance of each predictor calculated by the mean of the absolute value of regression coefficients derived by RLR. There are 100 columns in the heatmaps, with each column representing the coefficient values for one model. The performance of predicting A\(\beta\)-positivity in A\(\beta\)-negative individuals was higher with CSF-cohort compared to PET-cohort, suggesting the higher relevance of CSF data for conversion prediction. ADAS13:Alzheimer Disease Assessment Scale, 13 items, ADASQ4: ADAS Delayed Word Recall, MMSE: Mini-Mental State Examination score, RAVLT: Rey’s Auditory Verbal Learning Test, LDELTOTAL: Logical Memory Delayed Recall Total, TRABSCOR: Trail Making Test Part B, FAQ: Functional Assessment Questionnaire, CDRSB: Clinical Dementia Rating-Sum of Boxes, ICV: intracranial volume, MidTemp: middle temporal gyrus