Defining individualized networks

To facilitate group-level interpretations of individually-defined PFNs, we used a group consensus atlas from a previously published study in an independent dataset¹⁶ as an initialization for individualized network definition. In this way, we also ensured spatial correspondence across all subjects. This strategy has also been applied in other methods for individualized network definition^23,94. Details regarding the derivation of this group consensus atlas can be found in previous work¹⁶. Briefly, group-level decomposition was performed multiple times on a subset of randomly selected participants and the resulting decomposition results were fused to obtain one robust initialization that is highly reproducible. Next, inter-network similarity was calculated and normalized-cuts⁹⁵ based spectral clustering method was applied to group the PFNs into 17 clusters. For each cluster, the PFN with the highest overall similarity with all other PFNs within the same cluster was selected as the most representative. The resulting group-level network loading matrix V was transformed from fsaverage5 space to fslr space using Connectome Workbench⁹⁶, and thus the resultant matrix had 17 rows and 59,412 columns. Each row of this matrix represents a functional network, while each column represents the loadings of a given cortical vertex.

Using the previously-derived group consensus atlas¹⁶ as a prior to ensure inter-individual correspondence, we derived each individual’s specific network atlas using NMF based on the acquired group networks (17 × 59,412 loading matrix) as initialization and each individual’s specific fMRI times series. See ref. ²⁵ for optimization details. This procedure yielded a loading matrix V (17 × 59,412 matrix) for each participant, where each row is a PFN, each column is a vertex, and the value quantifies the extent each vertex belongs to each network. This probabilistic (soft) definition can be converted into discrete (hard) network definitions for display and comparison with other methods^19,23,94 by labeling each vertex according to its highest loading. Split-half reliability of the PFN loadings were assessed in ten participants who had the longest duration of low-motion quality resting-state data exceeding 20 min allowing us to derive PFNs in two 10-minute segments, as previously described in prior work⁹⁷, given the necessity of sufficient scan duration for the derivation of precision functional networks^20,21. This analysis revealed high intraclass correlation coefficients for PFN loadings across all 17 networks (ICCs: 0.84–0.99) indicating excellent reliability of this measure (Supplementary Fig. ⁴) that aligns with what has been found in prior work^20,21. For our univariate analyses, we calculated the total cortical representation of each PFN as the sum of network loadings across all vertices as previously¹⁶, using the soft parcellation to account for spatial overlap across functional brain networks. As described in prior work¹⁶, this measure of total cortical representation quantifies the spatial extent of each PFN on the cortical surface.