Applying Causal Discovery And Causal Hypothesis Testing To Single-Cell Transcriptomic Data

Biological systems,whether during development or disease progression,are driven and controlled by complex regulatory processes.Single-cell transcriptome sequencing technology is different from bulk sequencing.It can detect expression profiles of tens of thousands of genes in tens of thousands of cells at the single-cell level,instead of the average gene expression in all cells that masks biological signals.This creates an unprecedented opportunity for single-cell sequencing data analysis to infer the complex gene regulatory relationships within cells.Given that,most of the current analysis of single-cell transcriptomic data rarely examines gene regulatory relationships from the perspective of causality,analyzing single-cell transcriptomic data from the perspective of causality is of great significance to biomedical research.Graphical models play a crucial role in causality analysis.Graphical models(generally represented by directed acyclic graphs)can be thought of as representing the causal mechanisms behind data generation.The d-separation criterion can reveal the conditional independence relations among the variables contained in the graphical model,and these conditional independence relations can be tested with the conditional independence test algorithm.If these conditional independence relationships are disproved,then the corresponding graphical models are unlikely to represent true causal relationships.Using this property,it is possible to deduce the causal relationship from the data set(causal discovery),or to test a graphical model(causal hypothesis testing).The topic of this study is the use of graphical models to explore causal relationships from gene expression data.The contribution of this work can be categorized into three parts.First,this study proposes and develops a pipeline for inferring gene causal networks from single-cell transcriptomic data using causal discovery,then applies this pipeline to macrophage data in glioblastoma and lung cancer cell line data,and explains the reasonability of the results.Second,this study examines the principle,process and existing problems of causal hypothesis testing in detail,proposes solutions to some problems,integrates a variety of advanced conditional independence testing algorithms and develops several new composite testing algorithms.This study also develops an application which makes causal hypothesis testing easily applicable.Since the causal hypothesis test cannot be applied to the case of binary variables,this study also designed a binary variable causality discriminant method,and the results suggest that it outperforms the classical algorithm.Third,based on the KEGG pathways,this study uses causal hypothesis testing to help reveal the possible "activate" gene interaction mechanism behind the single-cell transcriptomic data of a certain phenotype,and explains the reasonability of the analysis results.The innovation and significance of this study include that it proposes solutions for the use of causal discovery and causal hypothesis testing to explore the causal mechanism behind biological data;analyzes and improve the existing methods of causal hypothesis testing;develops tools to facilitate researchers to use causal analysis for data mining;and designs a binary variable causality inference algorithm based on causal hypothesis testing.In conclusion,this study develops and integrates tools for causal analysis and apply these tools to some single-cell data for causality mining and develops corresponding tools,which is meaningful.These may greatly influence single-cell transcriptomic data analysis and promote biomedical researches.