Fig. 4: Receiver Operating Characteristic curves (ROCs) for logistic regression models predicting AD diagnosis in out-of-sample validation using the AddNeuroMed dataset (83 AD cases and 88 controls).
From: Cross-tissue analysis of blood and brain epigenome-wide association studies in Alzheimer’s disease
The training dataset included 135 AD cases and 356 control samples from the AIBL dataset. The best performing logistic regression model (with AUC = 0.696) included methylation risk score (MRS), age, sex, and estimated cell-type proportions, where MRS was computed as the sum of methylation beta values for 91 prioritized CpGs in cross-tissue analysis weighted by their estimated effect sizes in the AIBL dataset.
