Analysis Of Symptom Clusters Cross Different Diseases Based On Biclustering

The symptoms are important manifestations of disease diagnosis and treatment.Symptoms have specific rules in disease and patient performance.Previous studies have focused on the relationship between symptoms and disease.The main research objective is to find specific symptoms of disease.But for more than a decade,medical researchers have found that certain Symptom tend to occur simultaneously in different diseases and patients,forming Symptom Cluster,a Symptom Cluster with significant co-occurrence patterns.In view of the importance of symptom management,the identification of symptom groups and the discovery of their molecular mechanisms have become the hot topics in medical research.The traditional symptom cluster analysis method is mainly composed of principal component and hidden class model,and now the researchers mainly focus on the rules of symptom cluster under a specific disease.However,in the symptom cluster analysis of cross-disease species,there is often a need to identify the associated disease information,so there is a significant deficiency.Biclustering method was used to analyze the symptom group of cross-disease species.Using the data of disease symptom relationship with whole disease spectrum,the structure and molecular mechanism of specific symptom group were identified.The main research work of this paper includes the following two aspects:In order to study the rule of cross-disease symptom cluster,we first use the data from Disease ontology?Human phenotype ontology?OrphaNet?MalaCards surname He Unified Medical Language System A total of 16,383 high-quality disease-symptom relationships were collected and integrated in the library.The data include common diseases such as Diabetes,HIV,lung,heart,yellow fever,disorder,hepatitis chronic,Inflammation.In addition,there are symptoms such as Pain,fatigue,nausea,numbness,cough,vomiting,lack of appetite,weakness,distress.The data set contains a total of 13532 diseases and 2378 symptoms.Based on this data set,we use three classical Biclustering algorithms:BIMAX(Binary inclusion-maximal),QUBIC(Qualitative Biclustering)Cluster analysis of symptom groups by Spectral Clustering and Spectral Clustering revealed and identified many symptom groups with clinical significance.These symptom groups appeared simultaneously in different diseases,reflecting the common clinical phenotypes and patterns of different diseases.For example,there are three symptom clusters:pain,depression,and fatigue;Nausea and vomit;pain,fatigue,sleep disturbance,drowsiness and lack of appetite.they are also common concomitant symptoms in real clinical practice and do fit with real clinical experience.Based on this result,we combined 371422 symptom-gene relationships,with a total of 2834 symptoms and 16728 genetic data,and made use of 841068 interacting group data(from the large-scale integrated database STRING 11).The shared gene,shortest path and GO analysis were used to analyze.At the same time,we selected the same symptom group as the control group by random sampling and the results of the experiment.The experimental results show that these symptom groups obtained by Biclustering correlation algorithm have higher possibility of sharing genes and shorter molecular pathways of protein-protein interaction(compared with random symptom combinations),and have higher molecular mechanisms for sharing,P<0.05).Further comparative analysis shows that in the above three methods,we find that the QUBIC method has better biologically significant clustering results,while the spectral clustering has the best symptom clustering results in the shared molecular mechanism.In addition,the symptom clusters obtained based on Biclustering algorithm are in GO,pathway enrichment score of each annotation cluster is statistically significant,in the follow-up study,the performance of different methods for symptom group analysis can be further explored.