[Question] Testing Readfish on FAST5 file without playback

[harisankarsadasivan]

Hello, I'd like to measure the performance of ReadFish on FAST5 classification (basecall+minimap2) into target vs non-targets. I do not wish to do any playback. Just want to measure accuracy and performance on my GPU. Is there a way to do this?

[mattloose]

There is no way to do this within ReadFish without interacting with MinKNOW. ReadFish cannot take a fast5 file as input.

You can engineer something yourself using the fast5 api and guppy. You would need to work on the benchmarking carefully to ensure that you are not timing file access times and instead are benchmarking just the basecalling and alignment step. We did something similar (looking at accuracy and chunksizes) in figure 1 of this preprint: https://www.biorxiv.org/content/10.1101/2021.12.01.470722v2.full.pdf

A close approximation might be to look at the batch sizes and times as reported by readfish - but this won't allow you to ascertain accuracy - it will merely look at performance.

This isnt' something we can support through ReadFish.