Hardware requirements for NGS analysis

Question

[eb0906]

Our small bioinformatics core bought two Dell Precision T7610 Tower Workstations equipped with 1 Intel Xeon E5-2687W v2 Eight-core 3.4 GHz Turbo, 25 MB processor, 64 GB 1866MHz DDR3 RAM, 1GB NVIDIA Quadro K600 Video card, 256 GB Solid-state drive and two 1TB SATA drives, DVD-RW drive, 10Gb Network adapter, and an Nvidia Tesla K20C Computer Processor.

I am a novice user, but some initial thoughts I have are:

1. Do we have enough RAM to support multiple (2-3) RNA-seq analyses? For example, alignments, mapping, differential expression analysis, etc. Most of our work will involve RNA-seq, small RNA-seq, and ChIP-seq analysis.
2. Do we need an additional CPU? (Assuming we will be analyzing at least 2 RNA-seq experiments at any given time and will have additional users (2-3) logged on and trying to analyze their own data.)
3. It is my understanding that the greatest limiting factor in computational requirements for NGS analysis is I/O. At this point, is there any advantage to having a GPU versus CPU when it comes to NGS analysis?

Thanks in advance!

Answer 1

[Devon Ryan]

1. Depends on the aligner you and your users plan to use. If you have STAR in mind, then you might want more RAM for concurrent processes.
2. Probably, 8 cores is more appropriate for a single-user system.
3. Your understanding is correct and no, trying to use a GPU won't help you much (in fact, there isn't a lot that's GPU-based at the moment).

[eb0906]

Thanks, Devon.

For our purposes, we will primarily be using Bowtie. One of my colleagues is currently running cuffdiff on 16 c.elegans samples (15-16M reads/sample), and it looks as though it's stalled at the 'Processing Loci' step with 98% of the memory being used. This is our first attempt at using these workstations for an RNA-seq analysis, so we aren't sure what to expect.

[Devon Ryan]

That should suffice for a few instances of bowtie, it's slow and doesn't support spliced alignments (i.e., no aligning to the genome), but I guess it'll work.

The cufflinks issue doesn't surprise me, it's a really slow program.

Answer 2

[Damian Kao]

I am currently using a 64gb RAM machine with 32 cores. I find that's enough to do your standard mappings to genome/transcriptome and down-stream analysis.

However, that might not be enough if you plan to assemble transcriptomes or genomes. I've been using Amazon ec2 instances for a lot of my work. The high memory 244gb ram instances should be fine for most transcriptomes. I've also been using starcluster to setup ad hoc ec2 clusters and running abysss on that to assemble genomes. Here is one of my ipython notebooks on how to setup starcluster on ec2 and run abyss (http://nbviewer.ipython.org/github/damiankao/phaw-genome/blob/master/03_assembly.ipynb)

If you have the money, I highly recommend supplementing your hardware needs with AWS. I can usually find a high ram instance (r3.8xlarge) or high computation instance (c3.8xlarge) for ~35 cents an hour (spot instances) somewhere in EU West or US East.