A high-quality transcriptome is required to advance numerous bioinformatics workflows. Nevertheless, the effectuality of tools for de novo assembly and real precision assembled transcriptomes looks somewhat unexplored, particularly for non-model organisms with complicated (very long, heterozygous, polyploid) genomes. To disclose the performance of various transcriptome assembly programs, this study built 11 single assemblies and analyzed their performance on some significant reference-free and reference-based criteria. As well as to reconfirm the outputs of benchmarks, 55 BLAST were performed and compared using 11 constructed transcriptomes. Concisely, normalized benchmarking demonstrated that Velvet-Oases suffer from the worst results, while the EvidentialGene strategy can provide the most comprehensive and accurate transcriptome of Lilium ledebourii (Baker) Boiss. The BLAST results also confirmed the superiority of EvidentialGene, so it could capture even up to 59% more (than Velvet-Oases) unique gene hits. To promote assembly optimization, with the help of normalized benchmarking, PCA and AHC, it is emphasized that each metric can only provide part of the transcriptome status, and one should never settle for just a few evaluation criteria. This study supplies a framework for benchmarking and optimizing the efficiency of assembly approaches to analyze RNA-Seq data and reveals that selecting an inefficient assembly strategy might result in less identification of unique gene hits.
Keywords: de novo assembly; hybrid transcriptome; non-model organisms; normalized comparison; optimization; transcriptomics.