Prediction Method Of Microbial Interaction Based On Causal Analysis

Human gut microbes are a complex and dynamic ecosystem that play an important role in human health.The study of microbial time series data to infer the dynamic interactions between microbes is important to understand the mechanisms of cooperation and competition among microbes in the human gut.With the development of high-throughput sequencing technology,the time series data of massive microbiome has been made public,which has become the basis for systematic inference of cause-and-effect relationship between microorganisms.Based on the multiple linear regression model,the state-observation model and the maximum entropy estimation model,a causal inference method based on the time series data of microbiome is studied in this paper.The main contributions are as follows:First,a model combining the deep Boltzmann machine with multiple linear regression was proposed to infer the causal relationship between microorganisms.The deep Boltzmann machine is a probabilistic model composed of a random neural network.By combining the deep Boltzmann machine with the graph regularization vector regression model,the causal network obtained has better explanatory property and can play the role of screening subsets and clustering at the same time.The feasibility of the proposed method is verified by using the time series data of a group of microorganisms taking antibiotics as the experimental data and the predicted mean square error as the evaluation criterion.Secondly,a method based on the state-observed probabilistic model is proposed to infer the interactions between microorganisms.In parameter estimation,we introduce the Bayesian adaptive Lasso algorithm,which is a hierarchical probability model and can better solve the problem of data noise.Adaptive punishment is added to the model to constrain the parameters,so that the model can better combine the uncertainty with the cause-and-effect relationship between microorganisms,and the model has better interpretability,prediction ability and stability.Using the time series data of intestinal microorganisms taking antibiotics and the time series data of female vaginal microorganisms,and using the prediction mean square error as the evaluation standard,the proposed method has certain improvement in prediction accuracy compared with the existing method.Finally,a maximum entropy estimation model is proposed to infer the causal relationship of microbial time series.The parametric estimation model needs to establish the corresponding equation model according to the observed samples,which requires a large number of parameter estimation and is unstable.In this study,a parameter free estimation method,maximum entropy estimation model,was proposed to predict the cause-and-effect relationship between microorganisms.This method is based on the joint entropy formula,and the maximum point is obtained by linear optimization.Using the predicted mean square error,the maximum entropy method was found to be the best in the time series data of three samples of intestinal microorganisms taking antibiotics.The methods of parameter estimation and parametric estimation proposed in this study provide a new modeling method for inferring microbial causality based on time series data.