Rarefied OTU tables vs filtered OTU tables for group taxa summary plot?

[Restuadi Swatanto]

Hello,

First of all, Sorry, if I'm posting in the wrong forum. Actually, I'm still using QIIME 1.9 I'm new at this microbiome analysis works, so please forgive me if I asked many stupid questions.

My question is just a conceptual ones, hopefully someone can enlighten me a bit.

Here is a bit from core_diversity_analysis pipelines :
# Collapse samples in OTU table by categories command
collapse_samples.py -m map2_oct2017.txt -b table_mc10000.biom --output_biom_fp GROUP_otu_table.biom --output_mapping_fp GROUP_map.txt --collapse_fields 'GROUP'

# Sort OTU Table command
sort_otu_table.py -i GROUP_otu_table.biom -o GROUP_otu_table_sorted.biom

# Summarize Taxonomy command
summarize_taxa.py -i GROUP_otu_table_sorted.biom -o taxa_plots_GROUP/

# Plot Taxonomy Summary command
plot_taxa_summary.py -i GROUP_otu_table_sorted_L2.txt,GROUP_otu_table_sorted_L3.txt,GROUP_otu_table_sorted_L4.txt,GROUP_otu_table_sorted_L5.txt,GROUP_otu_table_sorted_L6.txt -o taxa_summary_plots/

# Group significance (GROUP) command
group_significance.py -i table_even10000.biom -m map2_oct2017.txt -c GROUP -o group_significance_GROUP.txt

The question is:
When plotting taxa for the group, why use the filtered Biom table (mc) instead of the rarefied ones (even)?

It would be much more fair to use the rarefied ones to compare between the group, I guess...

Thank you very much,

Restu

[Colin Brislawn]

Good morning Restu,

This is the perfect forum for questions about Qiime 1.9.1. Welcome!

I'll focus on your final question:

When plotting taxa for the group, why use the filtered Biom table (mc) instead of the rarefied ones (even)?

These two files have been processed in different ways.

The filtered biom table (mc2) only has OTUs with a Minimum Count of 2. This will remove less common OTUs and is a very common filter.

In the rarefied table, each sample has been subsampled to an even depth, so that each sample has the same number of reads in it. This is a simple way to control for bias due to sampling depth (because it throws away the extra depth).

Does that help answer your question?

Now I have a question for you:

Is it OK to process your data slightly differently in order to do a valid statistical test and in order to make a clear graph?

My answer to this question is 'yes,' and I often use my rarefied data for a stat test, then merge replicates and filter OTUs so that my graph more simple. What do you think?

Colin

[Restuadi Swatanto]

Thank you very much Colin,
This anwer is really helpful.

I guess, in my case, I'll do both, just to show that using mc and even is still not that different...

Thanks again