Research On Disease Prediction Method And Classification Based On Disease Network

With the help of artificial intelligence technology to analyze the occurrence and development mechanism of disease is the current research hotspot and difficulty.Massive clinical data provides opportunities for disease prediction and disease classification research.Disease prediction can provide help for the early diagnosis of patients,help to receive effective treatment in the early stage of the disease,reduce the pain,and relieve the economic burden.Deep mining of clinical diagnosis data of patients,identifying disease subtypes based on prognosis information of patients,and carrying out population difference analysis of subtypes play an important role in promoting individualized diagnosis and treatment of patients.Clinical studies show that there are phenotypic correlation and molecular correlation between a large number of comorbid diseases.The relationship between comorbid diseases provides a new idea for the prediction and classification of new diseases.In this paper,based on clinical data and molecular omics,and on the multi-source disease relationship network,we carried out the research of disease prediction and disease classification methods combined with deep learning,and finally improved the ability of auxiliary diagnosis and treatment decisionmaking.This paper is divided into the following two parts:(1)Research on disease prediction method based on disease networkThe existing research on disease prediction mainly focuses on the clinical disease information in the front page data of patients’ medical records,and fails to fully consider the relevant knowledge such as molecular omics.Firstly,by integrating high-quality multi-source disease relationship data,we constructed three disease networks with an average of 2183 diseases and 224 thousand relationships.Furthermore,we propose the disease prediction method DPNet-SD model and DPNet-DD model based on disease network.In the experiment,we initiatively takes the typical disease of severe pneumonia as the core,having constructed the disease prediction qualitative data set based on clinical sequential data and disease duration data.The experimental results show that the F1 value and AUC of DPNet-DD model are 0.8269 and 0.9037 respectively,and the performance of DPNet-DD model is significantly better than that of classical models(such as RF,SVM,etc.).It fully proves that the method we proposed can effectively predict the risk of patients with new severe pneumonia in the future.(2)Research on disease classification method based on heterogeneous networkThe prognosis of coronavirus disease patients is usually different.This study takes the coronavirus disease as the breakthrough point,combines the patient’s diagnosis and treatment records with multi-source medical data,and constructs a heterogeneous network with multi-source information.This study proposes a method of medical record information completion and disease subtype classification based on the disease network.A variety of network embedding representation methods are used to obtain node characteristics,and then the missing information in the patient’s electronic medical record can be reasonably completed.Finally,the groups of disease subtypes is divided by clustering method.The coronavirus disease based on Pro SNet was the best way to fill the information matrix.The survival analysis and dimension reduction visualization experiments showed that the best number of subtypes was 4.The 4 groups had statistic differences in disease,symptom and physical and chemical indicators.This paper provides a new idea for the classification of coronavirus disease.