I'm exploring a low-power edge device that detects pathogens / microbial variability in the field directly from nanopore raw signals (no basecalling). It would give very fast results using low power vs current GPU solutions.
I'm trying to get a reality check from practitioners.
Is sequencing itself still a bottleneck? Or has throughput/cost shifted the bottleneck to sample prep or library prep?
Would real-time detection (sub-second latency) provide clinical value beyond what batch basecalling + mapping gives?
What's the hardest part of your current workflow – sample prep, sequencing time, compute/analysis, or interpretation?
1 answer
Sample prep, and it isn't close. Getting enough pathogen DNA out of a messy field sample against a large excess of host background is where the time and the failures sit - by the time there's a library on the flowcell the analysis is the easy part. A device that saves compute is optimising a step nobody is really stuck on.
Worth checking the prior art before building, too. Raw signal mapping already exists in UNCALLED, Sigmap and RawHash, and adaptive sampling already makes real-time keep/reject calls on the instrument (ReadFish via basecalling, UNCALLED without it). So the novel part would have to be your power envelope specifically, not the skipping-basecalling part.
Can't help on the clinical latency question, I'm not on that side of it.
Log in to answer this question.