Fig. 6

Tau pathology spread modeling. a Predictions of log tau pathology from linear diffusion models based on bidirectional (anterograde and retrograde) anatomical connections. Solid lines represent the line of best fit, and shading represents 95% confidence intervals. Each dot represents a different brain region where tau pathology was measured. b Comparison of Pearson’s r values obtained by fitting bidirectional spread models using actual (black diamond) and alternate (purple points) seed regions. Using the true seeds yielded a better fit than using random seeds at both 3 and 6 MPI. c Comparison of Euclidean, anterograde, retrograde, and bidirectional model fits across 500 held-out samples derived from 500 iterations of a cross-validation process. Retrograde and bidirectional spread models performed significantly better than the anterograde and Euclidean spread models at 3 MPI, and the bidirectional model performed significantly better than all other models at 6 MPI (***P < 0.002)