Establishing a metagenomics lab/hardware requirements

Question

[iltheodorou]

Hello all!

I recently joined a lab that would like to set up a bioinformatics side of things. We do a lot of stool sample shotgun sequencing and we get loads of samples (in the thousands) sequenced to medium - deep sequencing (~10million reads per sample). At the moment we are still at the sequencing stage, so no analysis has started yet of the sequencing results.

For that we need to greatly upgrade our hardware. If one were to buy a workstation (instead of regular desktop or perhaps a server) what kind of specs would you suggest this workstation to have?

I am thinking the three things that are super important in decreasing order is RAM, space, and processing power.

For RAM, would you say for the amount of samples i referred to above, would below 100Gb be sufficient or should we be aiming for more?

For space, would you say tens of terabytes would be sufficient or should we aim already for hundreds of terabytes.

Finally, what kind of CPUs would you recommend? Would the intel ones be good enough i7 or i9? Is something else more state of the art this kind of analysis?

I have tried to do as much research as i can on my own but i am no bioinformatician! Any input would help :)

[Asaf]

It's not just storage capacity, it's also its speed.

[iltheodorou]

ok! thanks :) then what would you recommend?

[Asaf]

I don't know, you might want to find someone that established such an infrastructure lately. Do you have any support from your institute? A computer cluster, a backup system, most institutions have some of those things. You might want to reach out to the CS department if you have such or ask around labs that use HPC (astronomy?)

Answer 1

[Rahul Sharma]

That largely depends on type of analysis one wants to perform. For de-novo assemblies based on k-mer algorithms requires much RAM (for microbial genomes of size upto 200 Mbs my jobs consumed upto 512 Gbs of RAM). For routine bioinformatics analysis 124Gbs of RAM should be fine. For much computational intensive jobs one could use Google compute or other cloud based services instead of buying a new machine of ~12,000 EURs. Hope this will help you!

[iltheodorou]

Absolutely! We are mostly interested at the moment at taxonomic distribution between different samples and also on functional metagenomics, understanding what genes/pathways are present more/less in different samples. Not sure if these are in the computationally heavy categories or in the lower end of the spectrum.