Research On Imaging Genetics Association Analysis Based On Sparse Learning And Its Application

The increasing development of gene sequencing technology and imaging diagnostic techniques laid a technical foundation for imaging genetics.Imaging genetics,as an interdisciplinary field,its practical significance lies in studying the association between genes and brain imaging characteristics to explore the internal mechanism of the disease,and it plays an important role in mining biological information.Machine learning method is one of the main methods of imaging genetics research.We can learn from known biological data information and build correlation model based on sparse learning,and then solve the objective function to select risk genes and imaging features associated with disease.The risk factors we found can be used as an effective clue for disease diagnosis and targeted therapy.The main research work and innovation points of this thesis are as follows:Firstly,in order to effectively utilize the functional connection information between brain areas and complementary information of multimodal brain imaging data,this thesis proposes a regression method based on the fusion of structural and functional modal.Specifically,we adopt elastic network based on sparse learning algorithm to select the post-preprocessed sMRI and fMRI feature respectively and preserve the important voxels and functional network connectivity features.Then further multi-kernel support vector regression machine is used to predict the specific genotype.The experimental results on the public data set demonstrate the validity of the proposed framework,and the biological markers associated with the disease have been found in modal selection procedure.Secondly,recent research indicates that compared with the traditional construction method of network characteristics based on pairs nodes,the network construction method based on hypergraphs can reflect the function connection information between multiple brain regions,thereby transmitting the high order connection information.Therefore,this thesis also proposes a sparse multi-task correlation analysis method based on hypernetwork.In particular,we use the method of sparse representation to deal with fMRI data and build a hypernetwork,then extract three clustering coefficients from the hypernetwork as the functional characteristics of the brain.Lastly,the multi-task canonical correlation analysis based on sparse learning is performed to obtain the correlation between genotype and image features.The results show that the algorithm,based on the hypernetwork and multi-task,can enhance the correlation analysis effect,and also get some key biological factors that can help study the pathology of the disease.