Exploring the interactions among different brain regions about structure and function can better understand the pathological basis of neurological disorders.The brain functional network has been widely used in the diagnosis and classification of neurological diseases.However, because of the grouping effect among brain regions, there is some randomness in the hyperedge which is constructed by traditional method, and thus lacks the ability to explain the grouping effect information among brain regions and ultimately reduces the classification accuracy.So the method of constructing hyper-network was proposed and applied to the automatic diagnosis of people with Autism Spectrum Disorder (ASD) .Then a super network was constructed by the proposed method, which can perform group variable selecting based on pre-defined group variables replacing traditional single variable selection.The results showed that compared with traditional method of constructing hyper-network, the construction method based on Group Lasso can effectively remove the grouping effect among brain regions and improve the classification accuracy.