Which aligner for human WGS

Question

[quentin54520]

I have few human genome, it's paired end 151bp. I have 74 samples divided in 8 (4*2) fastq for each (592 files in total). I want to know which aligner is the best ? It's seem difficult to choose... I work on an hpc with this technical information:

Nodes 32 Physical cores 420 Maximum cores number per node 64 RAM 2,5 To Maximum RAM per node 256Go Graphical processors GPU Node 1 GPU 4 (Tesla K80) Cuda cores 19968 RAM 128 Go

Thanks a lot and sorry for this "general" question, it's the first time i do this...

[quentin54520]

Thanks a lot. And an other (maybe stupid) question : There is a real diffence (mostly in term of speed) beetween :

bwa mem .... > output.sam | samtools view... output.bam

Or

Bwa mem... output.sam

Samtools view ... output.bam

Rm output.sam

?

[ATpoint]

bwa mem idx fq1.fq fq2.| samtools sort -o sorted.bam

...is the most common syntax as this gives sorted files right away which you need for most downstream application. Don't direct files to disk with > if you pipe, that negates the purpose of a pipe.

I would just go with bwa mem, nobody will question that, it is arguably the most commonly used aligner and will in all likelihood serve you well.

Answer 1

[ATpoint]

There is no correct answer to this, basically all prominent aligners (bwa, bowtie2, bbmap, hisat2) are fine, but as many downstream tools, e.g. for variant calling and recommend in their manuals to use bwa mem I would go for this one. It does not make use of GPUs though (afaik none of the prominent aligners does, but I did not follow the developments recently, maybe something changed here).

https://github.com/lh3/bwa

[quentin54520]

I heard about soap3 but as i'm not a specialist maybe it a more reasonable choice to use a well establish bwa (what about bowtie2?) ?

Answer 2

[predeus]

I agree with the answer @ATpoint gave - bwa mem is an old and well-established tool for this task.

However, if you want to be very up-to-date, Isaac4 has shown the highest performance in latest benchmarks. It would probably be faster too.

[hermidalc]

Where can you find these benchmarks for Isaac4?

[predeus]

Sorry, hasn't logged in here in a while.

Here's our paper - includes Isaac4 among others: https://link.springer.com/article/10.1186/s12864-022-08365-3

Here's the analysis scripts: https://github.com/bioinf/caller_benchmark