Differential Sample-specific Network And Its Application To Individual Features

Complex diseases,i.e.cancer,have been a problem that has obsessed human for many years,threatening human life and safety.With the abundant and multi-dimensional datasets generated from cancer-related projects have presented great challenges and opportunities,which are contributed to unfold the complexity of diseases.Currently,gene chip technology is the most powerful tool for studying the human genome,and gene chip data analysis has become an important branch of bio informatics.By comparing the relationship between disease and gene expression regulation,the difference in expression profiles between normal tissues and cancer tissues,we can understand the molecular basis of diseases formation.Discovery of systematic and effective module biomarkers genes sets play an important role in personalized therapy and precision treatment of cancer,and sample-specific differential network(SSDN)is benefit to identify potential driver genes or modules in an individual patient.For construction of SSDN,we need to establish an SSN based on the previous conclusions.A reference sample set is required to construct a reference network in SSN,and the consistency of s-PCC needs to be considered in the SSN.The choice of reference network and the establishment of a difference network are crucial in research.However,there is no efficient way to assess whether the specific networks based on different standards have systematically correspondence,that means,whether the choice of reference network will affect the construction of different networks.In this study,we developed a statistical method,i.e.a sample-specific difference network(SSDN),to analyze the consistency of standard based on genes molecular expression of a single person.This paper theoretically confirms that the correlation of the reference network is consistency in the following two cases:1.The number of reference samples is very large;2.The reference samples follow the same distribution.Based on this theory,we generated simulated data and verified the consistency of the correlation,which also be verified by three gastric cancer progression from GEO datasets and three cancer datasets from TCGA database.Eventually,it can be verified that the real datasets were consistent by the theory.Therefore,when the number of samples of the reference network is very large,or the samples follow the same distribution,the correlation of the reference network is consistent.It provides a reliable theoretical support for determining the reference network in the subsequent construction of the difference network.Based on above theory,we established a disease-specific sample network(DSSN)for four cancer datasets in the TCGA database.We constructed the DSSN for each sample and our results demonstrate the application to characterize complex disease at the single-sample level.DSSN can also be used to identify patient-specific disease modules,hub genes as well as driver genes.Hub genes are highly correlated with individual somatic mutations,and have played an important role in cancer-related functional analysis.In particular,the results of survival analysis and the verification of ICGC database manifest these driver genes can be used independently as individual module bio marker,it futher illustrates the applicability of the proposed method.