Graph propagation network captures individual connectivity-function relationship through predicting functional activation based on structural connectivity

Brain regions’ functions are determined by the connectivity and establishing the connectivity-function relationship is an important goal in neuroscience. Previous researchers had successfully used functional connectivity to capture the individual differences in functional activations. However, establishing an individual connectivity-function relationship based on structural connectivity is more challenging and remains to be solved. To improve the prediction of individual functional activation based on structural connectivity, we proposed that multi-hop instead of direct structural connectivity should be used. We adopted graph propagation network which incorporates multi-hop structural connectivity features. We demonstrated that multi-hop structural connectivity improves the prediction of individual functional activations by over 90%. We also revealed intermediate brain regions along the multi-hop structural connectivity pathways that contribute to the individual prediction of functional activation. The results promote our understanding of how brain function emerges from structural connectivity network and contribute to the building of connectivity-function relationship.