Hello!  I kind of depends slightly on your downstream work flow. What do you want to do after you align the genome?  I don't know if there are many qualitative differences between tophat2 and STAR. I use STAR, but I have also used bowtie2 to do some processing with cufflinks afterwards.  STAR is recommended by the ENCODE project this was enough for me. I use STAR pretty much all the time, but I am doing differential expression experiments and I use HTSeq-count downstream followed by RUVSeq (edgeR). I know that you can set flags so that the STAR output can be used with cufflinks. If you are looking for differential splicing or non-canonical splicing events you can run STAR in 2-pass mode. It is also useful to use the HTSeq-count python scripts with DEXSeq if you want to look for differential exon usage. This could help you track down different isoforms of transcripts as well. I think that it may be easier to use the gencode genomes with STAR -genomeGenerate.  We use the ERCC controls and so we have to modify the gencode genome gtf file to include entries for the ERCC probes and we have to add the sequences to the gecode genome fasta file.  If I remember correctly when I was using bowtie I downloaded a prebuilt genome from them. Good luck!

STAR is robust against SNPs. Tophat2 has often problems to map reads with 2 or more differences (e.g. SNPs or seq. errors). Thus, you should definitely use STAR.

Oh and just because it might be interesting in the future. We have published a web service that allows you to fast benchmark a set of mapper on your genome and conditions. It is currently for genome sequencing, but we soon will release an RNA-Seq version were you can benchmark STAR and Tophat2 and others (e.g. Gsnap, HIsat) on the same data set.

Here is the link to our tool called TEASER: http://teaser.cibiv.univie.ac.at/

At the moment seems that tophat2 is going to be fully replace by HISAT2:

https://ccb.jhu.edu/software/hisat2/index.shtml

From the TopHat2 website:

TopHat 2.1.1 release 2/23/2016

Please note that TopHat has entered a low maintenance, low support stage as it is now largely superseded by HISAT2 which provides the same core functionality (i.e. spliced alignment of RNA-Seq reads), in a more accurate and much more efficient way.

For me: We still currently use TopHat2 > HTseq-count > EdgeR as pipeline to perform differentially expression analysis

Luca

Hi Martin,

I have done some testing and benchmarking with TH and STAR . I tried transcripts assembly after aligning them with TH and STAR separately. I used 4 assemblers in this case and most cases STAR is a winner. To back this up please find the table of stats attached.

Organism : Arabidopsis thaliana

Data: RNASeq stranded lib

Cheers 

Shab

If reference genome is available then i think TopHat2 have sufficient functionality to map spliced read. For significant mapping , you must know features of your genome like:-

- Minimum introne length

- Maximum introne length

- Polymorphism etc.

I have good experiences with STAR for mapping on a reference transcriptome. If you work with a reference genome and expect split read mapping to be important, I'd recommend to have a look into segemehl:

http://www.bioinf.uni-leipzig.de/Software/segemehl/

It runs slower, but in terms of accuracy it won't disappoint.

Best,

Rik

STAR is recommended by the ENCODE project, definitely should be your best option.

Thank you all for your comments.

Since most of the answers are leaning heavily towards STAR, i will use it for now.

My downstream goal is now finding differentially expressed genes and alternative splicing events.

Hi Martin,

It's a fast evolving field. You may consider using pseudo-alignment mappers as well.

The big challenge in quantitative analyses like RNAseq, is to distill your mappings into a table of counts per gene, transcript, or even isoforms. This is no problem for reads that unambiguously map to a single locus, but many mappings will have a certain degree of ambiguity. There are different ways to go about this. In the past I have combined STAR with RSEM, but this required some serious patching. I guess bowtie2 + RSEM is easier to use, and outcomes may only differ slightly. In the meantime I also explored pseudo-alignent (I used Kallisto) and that seemed to do a very good job (though I'm currently still comparing it to old-school mappings). Big advantage is the drastically reduced runtime...

Good luck!

Rik

Thank you Rik.

I will try Salmon soon and compare it with my STAR mappings.

Regrads!

STAR has a better mapping rate and faster than TopHat2. STAR has better sensitivity in variant detections (SNPs and INDELs) as compared to TopHat2. Hence STAR is recommended for RNA-seq data mapping. read more here https://www.reneshbedre.com/blog/star-aligner.html