Nanopore amplicon sequencing. I see, this is really metagenomics via Nanopore, an interesting approach.
Thats a different answer. You process via fastp but assemble via Megahit or possible 'Mega-Spades'. Then you filter the samples per isolate to obtain you 24 or 96 samples.
Everything will be full length - 'cause it's Nanopore. The analysis of this data is interesting. The project goal would be interesting to know, because it ain't normally how you'd do it and Nanopore is a bit of a luxury for straight amplicon sequencing (normally you'd use Illumina long read).
In summary the analytics side of this data needs a carefully derived hypothesis and it depends on your objectives, how many amplicons within an isolate you are loading into the sequencing machine and presumably its either 24 or 96 samples. There's potentially a lot analytics you can do with this data.
There are lots of approaches if this is a straight genotyping project to determine the presence/ absence of AMR - but it can get lost in isolation. Automating MLST typing can be done via phylogenetics and a type strain of each MLST type (separate question).
Here's my old response
This ain't quite the right question format, anyway you need Illumina. 24 vs 96 depends on the number of samples - I'd start with 24 if there's limited experience with your techs getting this working.
There are no singular pipelines for MRSA. I would recommend looking at Sharon Peacock's work as a starting point. You can perform MLST genotyping from the Illumina data along with AMR work.
Background Generally, this area of work is notably behind frontline population genomics in for example human evolution and even metagenomics. There has been some very cool modelling however describing the spatial distribution of MRSA, those papers are worth reading (I will not cite them). These papers are not Peacock's work BTW. The field has emerged from microbiology rather than statistical modelling and IMHO there remains a critical gap between population biology and phylogenetics that was - to my knowledge - never addressed. This will always hinder higher level inference. As an example I doubt any AI learning has been performed at the molecular level - whilst big chunks of this area of analytics are now rapidly becoming outdated methods.
Basic processing However, if you stick with regular processing via e.g. fastp (look it up if you don't know it) and then genome assembly. I strongly recommend de novo assembly. Please look at past posts here particularly concerning @Laura's questions for methods. Modern development of de novo assembly has been a fast evolving frontier in microbiology and you'll need to look a variety of methods and assess the 'best' (gives minimal alignment gaps).
This will provide the foundations for the two major banks of genotyping you seek AMR and MLST (if you've not heard of that look it up).
Future Once you've got to that point you can always post back for the analytics.
