Skip to main content

Table 2 Features and performance summary of the surveyed methods

From: Comparative assessment of methods for the computational inference of transcript isoform abundance from RNA-seq data

Method Extensive documentation Standard file formats Gene-level estimates Reconstruction supported DE analysis Efficient multi-threading Fast Small memory footprint
BitSeq X X    X X   
CEM   X   X    X X
Cufflinks X X X X X X   
eXpress X X    X    X
IsoEM   X X    X X  
MMSEQ X X X   X    X
RSEM X   X   X X   
rSeq X   X      X
Sailfish X X    X X X X
Scripture (X)* X   X     
TIGAR2 X X       
  1. To facilitate a user’s choice of method, we indicate which methods meet various criteria of usability, functionality, and performance, as follows: ‘Extensive documentation’ - documentation that goes beyond the description of parameters is provided (document, web page, FAQ which allowed us to run a given method confidently and without help from developers); ‘Standard file formats’ - the method exclusively operates on the indicated file formats for transcript sequences (FASTA), gene/transcript annotations (GFF/GTF or BED12), read sequences (FASTA or FASTQ), and read alignments (SAM/BAM as defined in [65] and produced by most modern aligners); ‘Gene-level estimates’ - estimates of expression on the gene level are provided in addition to those at transcript level; ‘Reconstruction supported’ - the method can also reconstruct transcript models based on the provided sequencing/alignment data; ‘DE analysis’ - the developers make a general recommendation or provide an integrated solution for differential analysis of transcript/isoform expression; ‘Efficient multi-threading’ - the method efficiently makes use of multiple cores (speedup of at least two-fold in at least three out of five datasets; see Additional file 2: Figure S2A); ‘Fast’ - processing of 100 million synthetic reads or their corresponding alignments completed in less than 1 h (16 cores and 64 gigabytes provided; see Fig. 1b); ‘Small memory footprint’ - all synthetic datasets could be processed with less than 8 gigabytes of memory (independent of the number of cores used; see Fig. 1c, d). Additional details are provided in the main text. *The documentation for the complete Scripture suite is extensive, but a detailed description of the archive ‘ScriptureScorer.jar’ that contains only the RNA-seq quantification module which we used here is not available. Furthermore, the options for this module are different from those described for the main program.