q2-repeat-rarefy: QIIME2 plugin for generating the average rarefied table for library size normalization using repeated rarefaction
- When handling a sparse dataset, I noticed that the rare taxa were easily ignored by the traditional one-shot rarefaction.
- To deal with this problem, I proposed the "Average Rarefied Table" method and wrote a very simple plugin (reference: q2-feature-table/_normalize.py at master · qiime2/q2-feature-table · GitHub)).
- Repeat rarefy simply runs random rarefaction N times, and computes the average count (floats are round up) of each OTU (ASV/feature) to generate the final average rarefied OTU table.
- It proves that comparing with the one-shot rarefaction, using repeat rarefy to normalize library size can keep significantly more OTUs (unpublished results).
- As the float average count of OTU is round up, the total OTU count of each sample may not be exactly the same.
- This method has the potential to be an ideal alternative to the current one-shot rarefaction, as it can keep information and avoid variation of composition.
- In addition to OTU (ASV/feature) table, the "Average Rarefied Table" method can also be extended to other profile tables (e.g., taxonomic profile table, gene profile table).
Installing
conda activate qiime2-2020.11
pip install git+https://github.com/yxia0125/q2-repeat-rarefy.git
Type "qiime repeat-rarefy" to test if the installation is successful.
Uninstalling
pip uninstall q2-repeat-rarefy
Using
qiime repeat-rarefy repeat-rarefy --i-table table.qza \
--p-sampling-depth 2000 \
--p-repeat-times 100 \
--o-rarefied-table average_rarefied_table.qza
The above example rarefied the 'table.qza', with the sampling depth of 2000 and the repeat times of 100, to 'average_rarefied_table.qza'.
You can set the sampling depth based on your own dataset and increase repeat times to 1,000, 10,000 ...
Citing
If you are interested to use this method, please include the following citation:
Yao Xia, q2-repeat-rarefy: QIIME2 plugin for generating the average rarefied table for library size normalization using repeated rarefaction, (2021), GitHub repository, https://github.com/yxia0125/q2-repeat-rarefy.