MOTIVATION
Long non-coding RNAs (lncRNAs) have recently gained interest, especially for their involvement in controlling several cell processes, but a full understanding of their role is lacking. Differential Expression (DE) analysis is one of the most important tasks in the analysis of RNA-seq data, since it potentially points out genes involved in the regulation of the condition under study. However, a classical analysis at gene level may disregard the role of Alternative Splicing (AS) in regulating cell conditions. This is the case, for example, when a given gene is expressed in all the different conditions, but the expressed isoform is significantly diverse in the different conditions (that is an isoform switch). A transcript level analysis may better shed light on this case, especially in studies having as goal, for example, a better understanding of the behavior of lncRNAs in lymphocytes T cells, which are fundamental in studies of specific diseases, such as cancer. After Cufflinks/Cuffdiff, several approaches for DE analysis at isoform/transcript level have been proposed. However, their results are often sensitive to the upstream analysis such as read mapping, transcript reconstruction and quantification, and it is often hard to choose "a priori" the most appropriate combination of tools. This work presents a tool for assisting the user in this choice, and poses the bases for a study devoted to the characterization of lncRNAs and the identification of of isoform switch events. Our tool includes a framework for the description and the execution of a set of DE pipelines over the same input dataset, as well a set of tools for reconciling and comparing the results.
METHOD
We designed an automated and easily customizable tool which is able to execute a set of existing pipelines for DE analysis at transcript level starting from RNA-seq data. Our method is built upon Snakemake, a workflow management system, with the specific goal of reducing the complexity of creating workflows. This approach guarantees that the experimentation is fully replicable and easy to customize. Each considered pipeline is structured in three steps: (i) transcript assembly, (ii) quantification, and (iii) DE analysis. By default, our tool builds and compares 9 different pipelines, each taking as input the same set of RNA-seq reads, obtained by combining different state-of-the-art methods to perform the transcript assembly (TA step) with different state-of-the-art methods to perform quantification and differential expression analysis (Q+DE step). More precisely, the 9 pipelines are obtained by combining two tools (Cufflinks and StringTie) and a Reference Annotation (Ensembl annotated transcripts) for the TA step, with three tools (Cuffquant+Cuffdiff, StringTie-B+Ballgown and Kallisto+Sleuth) for the Q+DE step.
Abstract truncated at 3,000 characters - the full version is available in the pdf file