Explaining discrepancies in microbiome composition caused by changes in sequence files and denoising methods.

[emankhalaf]

Hi @benjjneb

I am currently working with 16S sequences generated using the PacBio Sequel II platform. My dataset consists of approximately 700 sequence files, representing 16 genotypes. Each genotype includes sequence data from three different tissues. Given the large number of sequence files, the Pseudo option appeared to be the most practical approach for processing.

Additionally, I processed the sequence files for each genotype separately using the pooling option. In this case, I included sequence files from an additional tissue, resulting in each genotype comprising data from four tissues. My primary goal is to identify exact sequence matches of taxa across three specific tissues in each genotype.

When comparing results obtained using the Pseudo approach (applied to approximately 700 sequence files) with the pooling approach (applied to around 60–80 sequence files, including the additional tissue), I observed a reduction in the count of taxa with exact sequence matches across the targeted tissues in the pooling method. However, the overall taxa count for each tissue within each genotype remained relatively stable.

Additionally, I observed that some taxa disappeared from certain tissues when using the Pseudo option but were retained with the pooling method. This discrepancy appears to be influenced by the number of sequence files and the specific tissues included in each processing approach.

My question is: Is this discrepancy between the methods predictable? Additionally, could there be an explanation for the significant difference in the count of taxa with 100% sequence matches across tissues when using the two approaches? Why do some taxa disappear when changing the denoise method?

Your support is much appreciated!

Eman

[benjjneb]

There is some predictability in metrics like ASV richness when using different pooling approaches. A brief exposition of this is here: https://benjjneb.github.io/dada2/pseudo.html#pseudo-pooling

In general, a lot of these changes are happening amongst the low-prevalence/low-abundance fraction of ASVs.

[emankhalaf]

@benjjneb Thank you so much for your response. I observed that the reduction occurred in the count of a specific taxon representing a highly abundant species with multiple unique sequences (ASVs). This reduction was evident when the denoise parameter was adjusted to pseudo-pooling, using all sequence files from the project across all genotypes. Furthermore, the reduction in this taxon was more pronounced in a tissue that naturally exhibits high diversity for this group, whereas the reduction was minimal in another tissue with inherently lower microbiome diversity. I hope this explanation clarifies my point.

Notably, another factor that I believe could influence the inference of this taxon is the higher number of unique sequences observed when pooling the sequence files for each genotype after adding the sequence files from an additional tissue, known for its rich microbiome diversity.