We often perform alpha/beta diversity on OTU that created from subsampled data. Some other analysis, like taxonomy profile(abundance figure of different bacterial communities), PCA, RDA analysis, should them also be performed on subsampled data?
subsampled data means the number of sequences per samples was made equal with random resampling.
I recently asked a Professor a similar question!
I think it depends on the dataset. If every sample has a similar read count, using less computational steps or fewer methods can help to keep your analysis clearer. Additionally, if you subsample your data, the subsampled dataset will always be a little different and not 100% reproducible. (By its very nature.)
So if you have a similar read count, subsampling is not necessary.
If one of your samples has a much higher read count (10 or 100 times) it might seem more diverse than other samples, while not being more diverse. Here you can use rarefaction curves to compare samples with vastly different read counts and then argue that subsampling is necessary.
If you subsample, you have to explain that you subsampled and how you did it.
I think if you use subsampled data for one analysis and the complete dataset for another analysis, it might get complicated and confusing. Try to decide early in the analysis on what you want to show and what hypothesis you try to proof and then choose the methods accordingly.
I agree with Dr. Florian Prodinger, you should decide at the start of data analysis and it also depends on the variables you used, PCA is one of the most commonly used methods to show variations in microbial diversity in different samples
Huo San , when you refer to a subsampled OTU dataset are you taking about rarefaction? If yes, rarefaction curves are a good way to show that you were capable to "get" most of the diversity present in a specific sample. Check
Chapter Measuring and Estimating Species Richness, Species Diversity...
for more details.In my opinion, "subsampling or not subsampling", depends on the question.
- If you want to compare the principal component structures (PCA, RDA, etc) BETWEEN communities from the sample A, B and C, yes it is a good idea to use the rarefaction dataset. BUT, you need to decide a rarefaction cutting point!
- If you are interested in to understand the principal component structures WITHIN a specific sample, lets say sample A, then you DO NOT need to use the rarefaction dataset because you might lost predictive power.
Florian Prodinger in relation to the computational power. Indeed, many algorithms use a subsampling stepwise because is nonsense to calculate something with redundant data. Why to use the 100% of data when with the 20% you can get the same result? Regards,
- Badges
- Science topic
- Similar topics
- Bioinformatics
- Bioinformatics and Computational Biology