Study Of Meta-analysis Methods On Gene Expression Data

With the development and wide application of microarray and high-throughput sequencing technology,the amount of biological omics data is growing rapidly,including transcriptome.However,due to relatively small sample size,transcriptome studies usually have limited ability to obtain statistically effective results.To increase the reproducibility of transcriptome studies,meta-analysis has been performed to integrate gene expression data from multiple studies.Meta-analysis on gene expression data has great potential in clinical practice of disease management,especially discovering genetic markers for cancer diagnosis and prognosis.However,the overlap of cancer biomarkers from different studies is low,and the performance to predict clinical outcomes in independent validation cohorts is not good enough.In this thesis,we introduced more biological significance in meta-analysis to overcome the shortcomings of current meta-analysis methods.Using lung cancer as the example,we integrated gene expression data of lung cancer tissues from multiple studies and performed meta-analysis to discover gene signatures for lung cancer diagnosis and prognosis.First,we developed a pathway-based meta-analysis method and discovered gene signatures for lung cancer diagnosis and prognosis.Then,we explored a meta-analysis method to incorporate denoising autoencoder in gene expression signature identification and proposed a set of gene signature for lung cancer as well.Next,we performed meta-analysis on lung cancer gene expression data using existed methods,including the methods that using combined effect size and machine learning strategy.Finally,we compared the performance of different methods.We explored and established two new methods for gene expression data meta-analysis,which are based on pathway-level and deep learning strategy.Besides,we proposed new biomarkers for lung cancer prognosis and diagnosis.This study provides new thoughts for biomarker development from gene expression data and new strategy for meta-analysis on gene expression data.