I'm dealing for the first time with marker gene-based metagenomic analysis. Going through R tutorials and into literature I realised the issue whether counts needs to be rarefied to even depth has not been addressed yet. Many scientist support the need of rarefying to have real estimates of microbiome community while others highly discourage that as it is statistically inadmissible. Moreover, many software pipelines suggest to rarefy (i.e. Qiime) while others don't (philoseq R).
Should I rarefy to even depth before running alpha and beta diversity analysis, for differential abundances and eventually, to use picrust?
Danilo Basili, you should normalise your read counts across samples - otherwise samples with more reads will likely have higher alpha diversity. Rarefying should never be used for this purpose as its definition is 'random subsampling without replacement', which will results in biased data due to the random OTU picking. There are good alternatives to rarefying published. Just recently, we published an algorithm called 'scaling with ranked subsampling' (SRS) that is suitable for microbiome count data normalisation. SRS is implemented in R and easy to use:Article Improved normalization of species count data in ecology by s...
Feel free to sent me a message or reply to this answer if you have questions or comments.
Much success with your data, Lukas
Lukas Beule, At the time I raised the question I was quite a beginner in the analysis of microbiome data. Now I have a more complete understanding. Thanks for sharing the info about the SRS method, I'll have a look ;)
Dear Bernd, thanks for taking the time to answer this question. As you probably noticed the question was raised in 2017 when I was still a beginner with marker gene-based metagenomic analysis. Overall, the scientific community still accept the use of rarefaction to perform alpha and beta diversity analysis (as long as you don't use it for downstream analysis as differential abundance). Your suggestion to repeat rarefaction 1,000 times is good at least to improve the confidence in the diversity analysis results. However, it would be better to move towards more reliable approaches other than rarefaction as you and Lukas also mentioned.
Dear Danilo, nice topic and answers! I am not sure you have already seen it, but I recommend you to have a look at the following paper, which shows nice alternatives to normalize microbiome compositional data instead of rarefaction: https://www.frontiersin.org/articles/10.3389/fmicb.2017.02224/full
For instance, many researchers have preferred to use centered log-ratio (clr) transformation.
I hope it helps. Best of luck with your research!
- Badges
- Science topic
- Similar topics
- Filing