Efficient and Robust Transcriptome Reconstruction with StringTie

RNA-seq methods technologies are a key component in characterizing the structural complexity and expression profiles of the set of RNA transcripts produced in a cell. Using high-throughput sequencing machines, a single experiment can capture the expression levels of thousands of genes, some of them not included in standard annotation. The number of raw sequencing reads produced from one experiment can be enormous, typically numbering in the tens of millions even for a modest-sized assay. In highly expressed transcripts, the bundle of reads that come from the same gene locus in the genome can easily reach over 2 million reads. Therefore one of the main challenges posed by RNA-seq transcriptome assemly is a computational one, both in terms of memory, as well as speed, as it can be a challenge to store and analyze all the information needed to process all the read alignments: alignment starts and ends, cigar strings, strand information, mate pairs etc.

Recently we have developed StringTie, a genome-guided transcriptome assembler that runs much faster, and provides more accurate overall results than similar competing methods. Here we present new developments to our method, which reduce the big memory footprint usually associated to transcriptome assembly, while also keeping fast running times. Most RNA-seq runs can be processed by StringTie on a conventional desktop computer with as little as 8 GB of RAM, and even very large data sets run in under an hour.