What are the benefits of RNA-seq aligners that use GTF over just mapping to transcripts?

Question

[O.rka]

In particular, I want to compare STAR, BBMap, and BWA-MEM. "BBMap is a splice-aware global aligner for DNA and RNA sequencing reads." BBMap is my preferred mapper but I'm not sure how well it works for RNA-seq data of eukaryotic organisms with exons/introns.

Besides being faster, what benefit does STAR have over BBMap by incorporating a GTF file for RNA-seq?

[GenoMax]

BBMap is my preferred mapper but I'm not sure how well it works for RNA-seq data of eukaryotic organisms with exons/introns.

It should work as well as any splice-aware aligner out there.

STAR can generate gene counts directly and can allow you to skip a step. That is about the only advantage over BBMap I see in terms of functionality.

Even if STAR is faster than BBMap the difference is unlikely to be an order of magnitude, which is when I become concerned about time to completion component.

If you are looking for a fast option then use salmon with the transcriptome.

[h.mon]

You don't state what are the downstream analyses you intend to perform, nor what kind of sequences (short reads, long reads, something else?), both of which may be important when considering alternate mappers.

STAR manual states it will be more accurate when using an annotation, so STAR with annotation is better than STAR without annotation.

I don't know how this compares to BBMap, which doesn't use annotation information for mapping. As BBMap hasn't been published, there aren't many benchmarks with it included, but here is one: Best RNA-Seq aligner: A comparison of mapping tools, and here is another: STAR vs tophat2 - mapping reads to long exons. Take them with a grain of salt, as they are incomplete, short, and without detailed information.

BWA-MEM is not meant to be used for mapping RNAseq reads to the genome, but one could possibly use it to map to a transcriptome reference. In this case, reads that map to unannotated genes will either be unmapped, or worst, will map to a wrong transcript.