Deep Learning Based Cancer Genomic Data Clustering And Cancer Immunity Analysis

Cancer immunity is one of the hottest cancer research areas.Recent studies have shown that genomic alterations play an important role in cancer immunity.However,these studies are limited to the analysis of one single type of genomic alteration.Cancer is a complex disease whose pathological mechanisms are associated with multiple types of genomic alterations.Therefore,integrating multiple types of genomic alterations provides opportunities to comprehensively analyzing the impact of genomic alterations on cancer immunity.In this thesis,a deep learning model was developed to cluster cancer genomic data.Firstly,two genomic alterations data were collected from TCGA database,including somatic mutation genes(SMG)and microsatellite instability(MSI).Then,the functional alteration features were extracted from the raw data and encoded into a binary format.Finally,two deep belief networks were used to reduce the dimension of each type of genomic alteration features separately,and a deep Auto Encoder was used to cluster samples by integrating these genomic alteration features.7381 samples from 18 cancer types were clustered into three classes with distinct patterns of genomic alterations.By analyzing immune gene expression data,this thesis found one class with both high levels of SMG and MSI had high levels of immune gene expression,while another class with both low levels of MSI and SMG only had high levels of immune gene expression in 3 cancers.The other class with high levels of SMG but low levels of MSI had high levels of immune gene expression in 7 cancers.By applying the model to advanced melanoma patient samples who had undergone immunotherapy,this thesis found that the response rate was extremely high in the class with high levels of SMG but low level of MSI.These results indicate that these two genomic alterations can affect cancer immunity.The results of this paper not only reveal the association between genomic alterations and cancer immunity but also provide new insights into cancer genomics data related research.