Alzheimer’s disease (AD) is characterized by pathological aggregation of tau neurofibrillary tangles in the brain, most prominently in the temporal cortex. With the use of positron emission tomography (PET), the accumulation of tau can be spatially identified and quantified, resulting in highly informative imaging data of disease status and progression. However, tau-PET is expensive, invasive, and not readily available in most clinical settings. This project aims to evaluate the capacity of machine learning models to predict information extracted from tau-PET imaging using clinically available data, e.g. demographics (four features), MRI (294 features), cognitive tests (two features) and plasma biomarkers (six features). Data from the BioFINDER-2 cohort (n=1695) will be split into train (80%) and test (20%) sets. From the training dataset, ~150 different combinations of features, preprocessing steps, estimators and hyperparameters will be combined as possible pipelines in a machine learning regression model. The outcome variable will be a tau PET composite region previously found to provide optimal clinical value. The pipelines will be evaluated using 10-fold-cross validation, with aims of finding the model resulting in lowest mean absolute percentage error. The best models will be evaluated for the test set and compared to several pre-selected naive models which only include a subset of all available features. We have additionally identified multiple datasets with the same or similar features, which will be used for external validation.