See the accompanying code repository Track-Weighted Imaging (TWI) TW_FA_Gaussian25_155subj.mif.gz (dim 121 x 152 x 121 x 155): Track-Weighted Fractional Anisotropic images (TW-FA) generated with 25 mm neighborhood weighting (per-track smoothing) in template space. 155subj_FA_map.mif.gz (dim 128 x 128 x 88): Fractional Anistropic images (FA). wmfod_norm_up.mif.gz (dim 121 x 152 x 121 x 45): Study template derived from white matter fiber orientation distribution (FOD), normalized, and upsampled to 1.5 mm isotropic. SH_peaks_template.mif.gz (dim 121 x 152 x 121 x 9): Spherical Harmonics peaks derived from the study template (used to generate TOM and TOM trackings with TractSeg). Non-Negative Matrix Factorization (NMF) W.mat & H.mat: Solutions from NMF. The optimal solution found for this study is the one with 16 networks (i.e., in the 8th position in the mat files as there is a 2-step increment between solutions). Network_Parcellation_NNMF.nii: 3D volume containing the part-based solution derived from NMF (i.e., 16 spatially non-overlapping networks of structurally covarying white matter areas). Network_Parcellation_NNMF_cluster.nii: IDEM but only considering clusters of at least 25 connected voxels within each network (used to compute the overlap with TractSeg bundles). BRAIN VISUALIZATION For figure 2: TW_FA_Gaussian25_155subj_mean_mesh.obj: White Matter Mesh derived from the mean image (used with the surfice software). NMF_age_networks_composition.xlsx: Dataset containing the spatial, connectivity, and composite score used to determine the composition of each NMF-derived network. *_track_overlay.tck: The 6 Tractography files to overlay onto the mesh for visualizating the bundles most contributing to the middle-age-related NMF networks. For figure 3 & 4: TW_FA_Gaussian25_155subj_mean.nii: Mean TW-FA image across all subjects. TW_FA_Gaussian25_155subj_mean_mask_95.nii: Binary group mask containing only the voxels that had a non-null TW-FA value across at least 95% of subjects. *LC1_BSR_saliences.nii & *LC2_BSR_saliences.nii: Voxel-level salience map of the 1st and 2nd latent component (unthresholded; please note that only BSR +/- 2.58 signified a robust contribution in our study). NNMF_BSR_LC1.nii: Network-level salience map. *res & save_opts.mat: Partial Least Squares (PLS) results and parameters.