TransFlow: a modular framework for assembling and assessing accurate de novo transcriptomes in non-model organisms

Background: The advances in high-throughput sequencing technologies are allowing more and more de novo assembling of transcriptomes from many new organisms. Some degree of automation and evaluation is required to warrant reproducibility, repetitivity and the selection of the best possible transcriptome. Workflows and pipelines are becoming an absolute requirement for such a purpose, but the issue of assembling evaluation for de novo transcriptomes in organisms lacking a sequenced genome remains unsolved. An automated, reproducible and flexible framework called TransFlow to accomplish this task is described. Results: TransFlow with its five independent modules was designed to build different workflows depending on the nature of the original reads. This architecture enables different combinations of Illumina and Roche/454 sequencing data, and can be extended to other sequencing platforms. Its capabilities are illustrated with the selection of reliable plant reference transcriptomes and the assembling six transcriptomes (three case studies for grapevine leaves, olive tree pollen, and chestnut stem, and other three for haustorium, epiphytic structures and their combination for the phytopathogenic fungus Podosphaera xanthii). Arabidopsis and poplar transcriptomes revealed to be the best references. A common result regarding de novo assemblies is that Illumina paired-end reads of 100 nt in length assembled with OASES can provide reliable transcriptomes, while the contribution of longer reads is noticeable only when they complement a set of short, single-reads. Conclusions: TransFlow can handle up to 181 different assembling strategies. Evaluation based on principal component analyses allows its self-adaptation to different sets of reads to provide a suitable transcriptome for each combination of reads and assemblers. As a result, each case study has its own behaviour, prioritises evaluation parameters, and gives an objective and automated way for detecting the best transcriptome within a pool of them. Sequencing data type and quantity (preferably several hundred millions of 2 x 100 nt or longer), assemblers (OASES for Illumina, MIRA4 and EULER-SR reconciled with CAP3 for Roche/454) and strategy (preferably scaffolding with OASES, and probably merging with Roche/454 when available) arise as the most impacting factors.