Forum:GPU-enabled bioinformatics

Question

[Dave Carlson]

Hi all, it's been a few years since the topic of using GPUs in bioinformatics was last discussed. I'm putting together a list of common bioinformatics (loosely defined) workflows that can benefit from GPUs.

Here is what I've come up with so far (partially inspired by previous biostars conversations):

• Basecalling (e.g., dorado)
• Molecular Dynamics (e.g., AmberMD, Gromacs)
• Germline variant calling (e.g., Parabricks)
• Protein structure prediction (e.g., Alphafold, Openfold)

What are your favorite tools or general bioinformatics workloads that can benefit from GPUs?

Answer 1

[Mensur Dlakic]

I think you got the main applications covered.

It should be mentioned that fine-tuned protein LLMs can be used to predict almost any physical property of proteins (solubility) and many extraneous qualities (localization, optimal temperature for activity). There is also a generative side, where new proteins are created daily with improved properties.

https://github.com/RosettaCommons/RFdiffusion

After training and fine-tuning, DNA/RNA/protein sequences can be classified based on many different criteria. One of the questions that often gets asked here is identifying rRNAs in a mixture of sequences, and in my opinion the best tool for this task is GPU-dependent (it works without it, but very slow):

https://github.com/hzi-bifo/RiboDetector

Any field of biology/bioinformatics that deals with images is almost guaranteed to be heavily GPU-dependent. Same for many applications of machine learning that deal with large amounts of data, even if they are not images.

[Dave Carlson]

Thanks, Mensur! I was not familiar with RiboDetector, but it looks really useful.