Hey everyone we're looking at having an intern sketch out converting an algorithm to an FPGA and trying to work that into a Nextflow pipeline. I have not done any -omics myself directly for a while now and was wondering which nextflow pipeline (and which subcomponent of that pipeline) is the most important, popular and painfully slow? Ie if you had to pick one algorithm to move to an FPGA and therefore speed up by an order of magnitude, which would you choose and why?

Duplicate marking!

Duplicate marking is a good one, also search across a large collection of reads or variants is probably worth looking into. Long-read tools or spatial tools are a field that will probably see some increasing adoption still in the coming years. Generally, I think, most people in this forum are not tool developers themselves, so the intricacies of algorithms is usually beyond our interest and capabilities. The machines, such as Illumina sequencers, often make use of FPGAs, but that is basically proprietary software of the manufacturer. The only FPGA accelerated Nextflow pipeline that I am aware of is a proof-of-concept pipeline by Seqera with a local dragen module. I recommend to ask on the Biostars forum or possibly directly tool developers that you find on e.g. GitHub. Some Rust developers that come to my mind from the back of my head are Jianshu Zhao, Jim Shaw, Seth, Eike Steinig, Brent Pedersen, Noam Teyssier, Wouter De Coster, Michael Macias. Outside of Rust I have very little overview, but Rob Patro and Wei Shen come to my mind.

@Matthias De Smet @Matthias Zepper Thanks both. We are looking at sketching both of these out for a summer internship project.

Great news! Keep us posted on the results - I suppose your cloud will then soon provide machines with FPGA-acceleration for pipeline execution 🙂.

@Matthias De Smet @Matthias Zepper We found an eager intern willing to roll up his sleeves on this one. We just ordered an AMD FPGA and are getting the thing set up. Will share the repo with this group for sure 👊