I am a bit out of my field here, so not sure what is overall aim (identify bacteria in the samples or the phages?). I would probably put a bit more effort on the denoising step to get a better/more complete ASVs table to work with. Firstly using only R1 but you may consider also alternative strategies such as clustering on the reference (if you have it) with vsearch.

This is a case where I think going back to the moving pictures tutorial is going to help you. My recommendation is that you search the forum for topics related to rarefaction and picking a final depth. I will mention that your rarefaction curve is a tool to select a depth, and again, recommend the moving pictures tutorial as how to do this.

I disagree that spending more time with denoising is going to improve your results. ASVs are a major improvement over clustering (although its another good topic to search for if youâre interested) and resolution here is unlikely to improve for the sample. I guess you could try to increase your depth, but the flip-side is that youâre limited by having only 10 samples. And so, while you need more than 200 ASVs to work with for later characterisation, your detection limit for anything is also super low. No matter how much time you spend characterising the ASVs and tweaking your definitions, I doubt very much that youâll achieve statistical significance in feature based analysis for your experimental design.

Thanks @llenzi@jwdebelius
And @jwdebelius,I would really appreciaet your input on what type of problem you think is causing axes to be 11.11% for 5 axes and not making a 100% .

Thank you very much for your explanations; they are very helpful.

How rarefraction-depth and bray-curtis and other plots are related? I see rarefraction as an independent command to find the sampling depth, is there moe to it? I am sorry but just trying to get a good understanding.
And, @jwdebelius I did go through the moving tutorial again, thanks for your suggestions. That is where this question came to my mind.

I typically use the alpha-rarefaction plot to identify the rarefaction thresholds to use for the diversity analysis, with the diversity plug in (either phylogenetic or not). So it is the usual trade-off between coverage and number of samples. That may be also related on how many samples you have per experimental groups, you may afford to loose few samples across different groups, if you still retain enough samples per group to give a good statistical power to the analysis.

Thanks! I think @jwdebelius is more expert than me on eSamples and statistics in general, so I won't disagree with her! Only to clarify what I meant: maybe denoising using single reads would give you an abundance table less 'sparse' (with higher number of ASVs counted in more than 4 samples). Hence a possible better sequencing depth after denoising. If this in turn will give you a better statistical significance in feature based analysis I can not predict, although the small number of samples is surely limiting any statistical power for the analysis.