Modeling Of Dynamic Biological Network Based On Granular Calculation And Its Applications

With the development of biological big data,researches which focus on biological molecular mechanisms have made huge strides,benefiting from the perspective of data analysis.However,the large amount of high-throughput data also brings difficulties to the data analysis.In addition,the complexity of biological mechanism requires the use of more appropriate models to accurately describe the complex molecular interaction underlying the molecular mechanism.The network method is capable of describing the molecular interactions and delineating the molecular regulatory system in the form of a network.At the same time,granular computing theory is widely used to extract system structure,which is significant to understand how molecules achieve specific life processes through coordination at different levels.Based on the idea of coarse granulation and combining with the network analysis method,this article explores the molecular regulation mechanisms in dynamic progressions of diseases,mainly discussing the selection of predictive genes for lung adenocarcinoma,the development mechanism of type Ⅱ diabetes and the communications between circulating cancer cells(CTC),nature killer cells(NK cells)and platelets during the blood metastasis of colon cancer.The main contents are as follows:In the second chapter,in order to extract several genes as predictors to classify tumors vs normal samples,a novel approach was proposed based on Granger Causality test and stepwise character selection,with the object of maximizing classification precision and minimizing number of predictors.Firstly,based on the data of methylation,gene expression and miRNA expression,we constructed the gene-gene interaction network(GGN)and then the diff-genes were obtained by analyzing differential expression.Furthermore,the feature-genes were identified by network degree analysis.Finally,Granger Causality test and Pearson correlation test based on the interaction network were utilized to remove “dependent-genes” from feature-gene set,and a stepwise character selection algorithm based on Random Forest classification model was further constructed.In our experiment,only 6 genes were exacted as resulting predictor genes,including TOP2 A,GRK5,SIRT7,MCM7,EGFR,COL1A2.Robustness of this approach was validated by applying this 6-predictor-model into 6 independent datasets.High precisions(ranged from 95.3% to 100%)indicated that our method was useful to classify patients and healthy individuals,which was also helpful in shortening the diagnosis time in clinical medicine.In the third chapter,the dynamic evolution of type Ⅱ diabetes(T2D)in mice was explored based on the time series gene expression data.Here,we develop an updated computational framework named after VD-analysis.In dynamic network methods,the disease progression can be analogous to an animated film composed of discrete frames,where each frame represents a temporary state of the time-varying gene-gene interaction network.The major shortage therein is that the transition between two neighboring temporary states was beyond investigation.The proposed VD-analysis framework improved dynamic network method,which identified modules in each transient GGN and then measured the transition relationships between the modules involved in adjacent transient GGN,so that to delineate the underlying molecular mechanism that drive T2 D progress.In addition,we firstly introduce V-structure — a gene module composed of three genes and two interactions among them — and define it as unit module.Result indicates that the whole process of T2 D is exactly divided into 3 stages,namely,early stage,transition stage and later stage,and that the driver V-structures inferred for each stage are qualified biomarkers.In summary,our method contributes to the description of dynamic disease progression and the V-structure biomarkers facilitate the treatments of disease.In the fourth chapter,the communications mechanism among CTC,NK cells and platelets during the metastasis of colon cancer was studied,by considering both the intercellular molecular interaction network and the intercellular communication.Firstly,intracellular GGN for each cell was constructed,according to the differences of gene expression levels in cells.Secondly,the ligand-receptor interaction was used to connect the two intracellular networks to obtain the intercellular communication network.Furthermore,we improved the module extraction algorithm based on hierarchical clustering and then applied it to identify the functional modules in the intracellular network.In addition,the correlation levels between the modules in the two cells was calculated by random walk algorithm.Finally,we found modules responsible for releasing signals and receiving signals,respectively.Several key ligand-receptor pairs that significant in signal transition were extracted as well.We also identified 4 predictor genes for distinguishing metastatic samples of colon cancer and the normal samples,via the stepwise feature selection algorithm.The resultant predictor genes including LIMK2,ARHGEF6,F2RL1 and ITGA8,with ACC = 98.42%.Moreover,ROC curve indicates the efficiency of the classification model(AUC=0.78).In summary,considering the integrity of molecular regulatory mechanism,it is of great significance to analyze the intercellular and intracellular molecular interaction in the study of intercellular functional synergy in the process of cancer metastasis.