RNA-seq analyses: Benchmarking differential expression analyses tools reveals the effect of higher number of replicates on performance

Abstract

16The introduction of several differential gene expression analysis tools has made it difficult for 17 researchers to settle on a particular tool for RNA-seq analysis. This coupled with the appropriate 18 determination of biological replicates to give an optimum representation of the study population 19 and make biological sense. To address these challenges, we performed a survey of 8 tools used 20 for differential expression in RNA-seq analysis. We simulated 39 different datasets (from 10 to 21 200 replicates, at an interval of 5) using compcodeR with a maximum of 100 replicates. Our goal 22 was to determine the effect of varying the number of replicates on the performance (F1-score, 23 recall and precision) of the tools. EBSeq and edgeR-glmRT recorded the highest (0.9385) and 24 lowest (0.6505) average F1-score across all replicates, respectively. We also performed a 25 pairwise comparison of all the tools to determine their concordance with each other in 26 identifying differentially expressed genes. We found the greatest concordance to be between 27 limma voom treat and limma voom ebayes. Finally, we recommend employing edgeR-glmRT for 28 RNA-seq experiments involving 10-50 replicates and edgeR-glmQLF for studies with 55 to 200 29 replicates. 30Author summary 31 Downstream analysis of RNA-seq data in R often poses several challenges to researchers as it is 32 a daunting task to choose a specific differential expression analysis tool over another. 33Researchers also find it challenging to determine the number (replicates) of samples to use in 34 order to give comparable and accurate results. In this paper, we surveyed eight differential 35 expression analysis tools using different number of replicates of simulated RNA-seq count data. 36 We measured the performance of each tool and based on the recorded F1-scores, recall and 37 precision, we made the following recommendations; consider edgeR-glmRT and edgeR-glmQLF 38 for replicates of 10-50 and 55-200 respectively. 39 40 42 exponential increase in RNA-seq data generation with an equivalent rise in the development of 43 algorithms for differential gene expression (DGE) analyses with varying performances. These 44 methods seek to make data analyses relatively easier and address complex biological questions 45 with greater levels of statistical confidence. However, the challenge still remains the selection of 46 optimal DGE tools and sample size calculations for optimal accuracy. This makes the selection 47 of tools and sample sizes for optimum analyses a very crucial but daunting task. 48Over the years, several research articles have been published that address the lack of consensus 49 among DGE tools. Examples of these are the works of Seyednasrollah et al (1) who performed a 50 systematic comparison of some popular DGE tools and provided recommendations for choosing 51 the optimal tool. Rapaport et al (2) assessed a number of tools based on the performance of 52 normalization, false-positive rates and the effect of sequencing depth and sample ...

Comprehensive analysis of Arabidopsis thaliana DNA polymerase epsilon catalytic subunit A and B mutants – an insight into differentially expressed genes and protein-protein interactions

“…Several different bioinformatics tools have been developed and used in the analysis of expression data generated through next-generation technologies. However, having multiple tools for analysis has made the analysis and interpretation of data challenging for researchers and has led to large variations in outputs [20, 21]. Hence, in the present study, two computational tools for transcript assembly (Cufflinks and HTSeq) and three tools for differential gene expression analysis (cuffdiff, edgeR and DESeq) were used; only DEGs resulted from all the tools were considered for the downstream analysis.…”

Critical Differential Expression Assessment for Individual Bulk RNA-Seq Projects

“…Several bulk RNA-Seq differential expression benchmarks have been previously published [1][2][3][4][5][6][7][8]. However, if data analysis is performed for a sufficient number of comparisons, then we expect that some methods troubleshooting may be needed.…”