A New Method For Pathway Analysis And Its Application In Metabolomics

Metabolic pathways are metabolic modules with specific functions.The pathways are highly connected with each other to form a complex metabolic reaction network,which cooperates to complete the metabolic function of the organism.In disease metabolomics research,the discovery of disease-specific metabolic pathways helps us better understand and study the pathogenesis of diseases.From the perspective of systems biology,many metabolites participate in more than one pathway,and in biological systems,pathways are closely related and work together.For different pathways,their sizes(that is,the number of metabolites in the pathway)and pathway the dependencies among the internal metabolites are very different,which is also a challenge accurately identifying the relevant metabolic pathways.Aiming at these problems,this paper mainly completed the following research work:(1)A pathway analysis method based on overlapping metabolite sets is proposed.This method builds a multivariate model called og PLS on the entire data set.The weight vector of the model is decomposed into pathway-specific sub-vectors according to the membership between pathways and metabolites.The sub-vectors are independently updated during the calculation process.An optimized debiasing coefficient is introduced to each pathway for the certain comparability between pathways.Finally,the pathway selection in the model is realized by combining group lasso.This paper evaluated the performance of the proposed method using the simulated data and compared it with Gobal-test and MB-PLS-PIP methods.The results show that this method,although more computationally complex,outperforms the others in accuracy,robustness and reliability.(2)For real metabolomics datasets,the number of perturbed pathways is unpredictable.In the og PLS model,the value of the sparse coefficient directly determines the number of selected pathways.If the sparse coefficient is too large or too small,the results of the path analysis will be greatly deviated.Based on this,this paper improved the og PLS model based pathway analysis method,and adopted a stability selection strategy to rank the importance of pathways,so as to avoid multiple selection or missed selection of pathways caused by selecting a specific number of pathways.In addition,through the simulation study,this study found that the results of pathway rankings in this method are not sensitive to model parameters,which avoids the problem of model parameter optimization to a certain extent.This paper applied this method on a colorectal cancer metabolomics data,analyzed the differences between cancer group and healthy group,cancer group and polyp group,and compared with Glabal-test and MB-PLS-PIP two pathway analysis methods.The results show that the overall trends of pathway rankings in the three methods are basically the same,but the top ranking pathways in this method have higher pathway independence and cumulative metabolite coverage,and lower cumulative metabolite overlap rate,so the results have higher reliability.