MultCausal:A Unified Method For Causal Gene Inference Using Multiomics Data

The inference of causal gene is significant in both theory and practice,including physiological processes discovery,especially the mechanism of human diseases,crop improvement and disease treatment.Therefore,the development of statistical methods for infering causal gene is very important.The existing causal gene inference methods can be dived into three categories: genomic-data-based method,other-omics-data-based and multiomics-data-based method,and network-based method.Many methods and many improvements have been come up with,but there are still some issues here.In one hand,these methods cannot use multiomics data formally at the same time,in the other hand,these methods are not isolated,but the relationship between these methods has not been clearly explained.In view of this,a causal gene inference method named MultCausal has been developed in this paper.The main idea is defining design matrix Z by the matrix multiplications of multiomics data,then put matrix Z into linear model and solve it,and log-transform the P value of each gene.The relative values of them will be the relative chance of these genes being casual gene.To validate the effectiveness of this method,this paper firstly chose one method from each category(linear regression,GSMR,GeneRank),and compared them with MultCausal in simulated dataset and Arabidopsis Flowering Time dataset.Compared with the other three methods,MultCausal performed better.It can control the false positive rate efficiently while keep high statistical power.Then this paper discussed the relationship between MultCausal and the other three methods,and proved the equivalence between MultCausal and linear model,the equivalence between MultCausal and GSMR formally,and proved the relevance between MultCausal and GeneRank using simulation.Moreover,this paper imporved the performance of MultCausal by turning Z into f(Z) inspired by SVR.