Figure 9

Example scenario in which RSEM-EVAL correctly selects the true assembly whereas Genovo and ALE select suboptimal assemblies. Because Genovo and ALE do not explicitly take into account transcript abundance and read mapping uncertainty, scenarios in which multiple isoforms of the same gene are present in an RNA-Seq sample can confuse these methods. In this example, a gene has two isoforms, the first isoform (with a length of 1,000 bases) corresponding to the first half of the second isoform (with a length of 2,000 bases). We simulated 5,000 single-end RNA-Seq reads of length 100 bases with 0.01% sequencing error from these transcripts and with a 90:10 abundance ratio between the first and second isoforms, respectively. Because RSEM-EVAL models transcript abundances and takes into account read mapping uncertainty, it correctly scores the true assembly the highest. In contrast, Genovo selects the assembly containing only the long isoform and ALE selects the assembly containing only the short isoform.