Hi everyone,

I am analyzing shotgun metagenomic sequencing data from human fecal samples generated using Oxford Nanopore (ONT) long-read sequencing.

I have a question regarding preprocessing (QC and host removal) strategies for ONT reads when performing two downstream analyses:

• Read-based taxonomic profiling (Kraken2 or nf-core/taxprofiler)
• Metagenome assembly and binning (nf-core/mag)

From my understanding, these two approaches optimize for different objectives:

• Read-based profiling prioritizes accuracy at the individual read level and may benefit from stricter filtering (higher Q-score and length thresholds) to reduce misclassification errors.
• Assembly-based analysis prioritizes coverage and genome recovery and may benefit from retaining more reads, including lower-quality ones, to improve contiguity and completeness.

Given this, I would like to ask:

Is it considered good practice to apply different QC strategies to the same raw ONT data depending on the downstream analysis (like running nf-core/mag and nf-core/taxprofiler separately with different filtering thresholds)?

For example:

• More stringent filtering for read-based taxonomic profiling
• More permissive filtering for assembly and binning

In particular, after adapter trimming and host removal, I am unsure whether additional read filtering (e.g., using Filtlong) should be:

(A) avoided, and the same cleaned reads used for both workflows, or

(B) applied with different parameter settings depending on downstream analysis, for example:

taxonomic profiling:

filtlong --min_length 500 --keep_percent 95 --min_mean_q 7

assembly / binning:

filtlong --min_length 1000 --keep_percent 90 --min_mean_q 10

Additionally, if Filtlong is used, what is the recommended order relative to host removal? Should read filtering be applied before host removal (e.g., Filtlong -> minimap2), or after host removal (e.g., minimap2 -> Filtlong), especially in the context of human fecal metagenomic data?

Or is it generally recommended to perform a single unified preprocessing step (QC + host removal, with or without Filtlong) and reuse the same cleaned reads for both pipelines?

In other words, what is the current best practice in real-world ONT metagenomic workflows: a single shared dataset vs analysis-specific filtering after host removal?

Any insights or references would be greatly appreciated. Thank you in advance.