Comprehensive Analysis of the Influence of Technical and Biological Variations on De Novo Assembly of RNA-Seq Datasets

Abstract

De novo assembly of transcriptomes from species without reference genome remains a common problem in functional genomics. While methods and algorithms for transcriptome assembly are continually being developed and published, the quality of de novo assemblies using short reads depends on the complexity of the transcriptome and is limited by several types of errors. One problem to overcome is the research gap regarding the best method to use in each study to obtain high-quality de novo assembly. Currently, there are no established protocols for solving the assembly problem considering the transcriptome complexity. In addition, the accuracy of quality metrics used to evaluate assemblies remains unclear. In this study, we investigate and discuss how different variables accounting for the complexity of RNA-Seq data influence assembly results independently of the software used. For this purpose, we simulated transcriptomic short-read sequence datasets from high-quality full-length predicted transcript models with varying degrees of complexity. Subsequently, we conducted de novo assemblies using different assembly programs, and compared and classified the results using both reference-dependent and independent metrics. These metrics were assessed both individually and combined through multivariate analysis. The degree of alternative splicing and the fragment size of the paired-end reads were identified as the variables with the greatest influence on the assembly results. Moreover, read length and fragment size had different influences on the reconstruction of longer and shorter transcripts. These results underscore the importance of understanding the composition of the transcriptome under study, and making experimental design decisions related to the need to work with reads and fragments of different sizes. In addition, the choice of assembly software will positively impact the final assembly outcome. This selection will affect the completeness of represented genes and assembled isoforms, as well as contribute to error reduction.