search

readwrite112 / AGAThA

PPoPP24 AGAThA: Fast and Efficient GPU Acceleration of Guided Sequence Alignment for Long Read Mapping
13 stars 1 forks source link

Using AGAThA for accelerating alignment of large sequencing datasets #4

Open kisarur opened 4 months ago

kisarur commented 4 months ago

We routinely use Minimap2 for processing large nanopore sequencing datasets which take hours. For example, aligning a single 30x dataset to the human genome using Minimap2 takes approximately 12 hours. I read your paper and the speed ups are amazing. A 18x speedup means we could reduce a 12h execution to 40 minutes! Before trying it out, could I know if the reported times are end to end? That is, from input FASTQ sequences to an output SAM file?

Thank you very much!

readwrite112 commented 4 months ago

Thank you for taking an interest in our work. AGAThA focuses only on accelerating the 'Extend' part in Minimap2. Thus, the reported time in the paper is limited to that part and does not cover the entire end-to-end process.

We are currently working on extending this kernel to a full end-to-end acceleration of Minimap2, but it's difficult to give a specific timeline at the moment.

(In case you're still interested in reviewing the code) I'm currently modifying some parts of the code and the documentation, and I strongly recommend revisiting the code after these are done (I've found some issues here and there).