benjjneb
commented
3 years ago As you can see, we lost ASVs with pooled1 and pooled2 methods compared to default method. But with pooled2 method we lost an abundant ASV (53793 reads) which is present with pooled1 method. I don’t understand why, because in pooled2 method I set the minFoldParentOverAbundance to 8. Do you have any idea?
I don't understand how this could happen. Are pooled1 and pooled2 operating on the same dada(..., pool=TRUE) output? Just a thought, but could there be a typo causing the `minFoldParentOverAbundance to not be set as expected? I see one in the command in your post in pooled2.
Based on these results, can we say that pooled1 method will be a good compromise and therefore the parameter set to choose for these analyses.
On these rough numbers, they all look reasonable. I would want to clear up the abundant ASV issue though, and choose one of the methods that is keeping it (probably).
My second question is related to how to handle several runs of different sites in the chimera removal step. I have multiple runs for one site and multiple sites. Since I then want to compare the sites to each other, I am wondering if I should merge all runs for one site and do the chimera removal step per site or merge all runs for all sites together and do the chimera removal step then. What do you think?
Merge runs first, then do chimera removal after. This keeps chimera removal consistent across runs.
benalric
Dear Benjamin,
I work with time series datasets of metagenomic data in several sites. For each site, I have several runs.
I have a first question about what setting parameters to use in step to infer sample composition and in step to remove chimeras. Following the information in discussions #218, #887, #1042, I run several analyses with several combination of parameters: i. default: dada(…, pool = FALSE) and removeBimeraDenovo(…, method = “consensus”) ii. pooled1: dada(…, pool = TRUE) and removeBimeraDenovo(…, method = “pooled”, minFoldParentOverAbundance = 2) iii. pooled2: dada(…, pool = TRUEE) and removeBimeraDenovo(…, method = “pooled”, minFoldParanetOVerAbundance = 8) I made taxonomic assignation with the naive Bayesan classifer methods proposed in DADA2.
Combination | ASV | read | max_read | Kingdom | Species | identity
default | 735 | 3152227 | 107730 | Eukaryota | Heterocapsa_pygmaea | 100 pooled1 | 679 | 3133661 | 107730 | Eukaryota | Heterocapsa_pygmaea | 100 pooled2 | 715 | 2925222 | 107730 | Eukaryota | Heterocapsa_pygmaea | 100 diff_d_p1 | 65 | 27841 | 6308 | Eukaryota | Gyrodinium_spirale | 81 diff_d_p2 | 34 | 75363 | 53793 | Eukaryota | Gyrodinium_helveticum | 81 diff_p1_d | 9 | 15079 | 9954 | Eukaryota | Chaetoceros_sporotruncatus | 81 diff_p1_p2 | 9 | 65993 | 53793 | Eukaryota | Gyrodinium_helveticum | 83 diff_p2_d | 14 | 19488 | 9954 | Eukaryota | Chaetoceros_sporotruncatus | 80 diff_p2_p1 | 45 | 22880 | 6308 | Eukaryota | Gyrodinium_spirale | 87
In the table, ASV: number of ASVs, read: number of reads, max_read: number of reads for the most abundant ASV, identity: identity level after taxonomic assignation As you can see, we lost ASVs with pooled1 and pooled2 methods compared to default method. But with pooled2 method we lost an abundant ASV (53793 reads) which is present with pooled1 method. I don’t understand why, because in pooled2 method I set the minFoldParentOverAbundance to 8. Do you have any idea? Based on these results, can we say that pooled1 method will be a good compromise and therefore the parameter set to choose for these analyses.
My second question is related to how to handle several runs of different sites in the chimera removal step. I have multiple runs for one site and multiple sites. Since I then want to compare the sites to each other, I am wondering if I should merge all runs for one site and do the chimera removal step per site or merge all runs for all sites together and do the chimera removal step then. What do you think?
Thank you in advance for your answer.
All the best,
Benjamin