Good afternoon,
Can anybody advise me to what kind of transformation (squareroot, forthroot or none) should I do to observe microbial patterns distribution in different conditions?
Thank you,
Vânia
Hello for microbial distribution depends from number of colonies in plate you can use Poisson distribution or log normal distribution.
1.The Poisson distribution is used to describe the distribution of rare events in a large population. For example, at any particular time, there is a certain probability that a particular cell within a large population of cells will acquire a mutation.
2. Log-normal distributions are usually characterized in terms of the log-transformed variable, using as parameters the expected value, or mean, of its distribution, and the standard deviation. This characterization can be advantageous as, by definition, log-normal distributions are symmetrical again at the log level.
Good luck!
Can you clarify whether you are referring to plate counts or sequencing data?
Assuming you are referring to sequencing data-
Read up on analysis of compositional datasets- "A Concise Guide to Compositional Data Analysis John Aitchison" and consider using a log-ratio transformation to deal with the sparsity. The Primer6 software does not have these corrections in-built, so you should deal with the data externally and import as a distance matrix.
The use of transformation before ordination methods such as nMDS, is partly to reduce the weight of outlier samples. In other case would be to allow comparison between variables with different units.
One way to assess if you need to perform any transformation is to do a Draftsman plot in PRIMER (Analyse, Draftsman Plot). It will plot the variables against each other and you can assess if the data are squeezed against one axis or are evenly distributed. If the data are against one axis they probably need transformation. In general, if you have large differences between samples, transforming your data will not "hurt". You can always run the nMDS without transformation and assess if the 2D stress of the MDS is good (i.e. < 0.2 at least), or not, and if clearly you have outliers or not.
For the choice of transformation, basically each method have a different strength, from "weak" to "strong" transformation as follow: square root < square root < fourth root < log.
Note that if your data include zero, for log you need to do log (x+1).
Adam Wyness I'm working with sequencing data. Thank you for your answer!
Aimeric Blaud Thank your for your answer. I did the Draftsman Plot and it looked evenly distributed (however i'm working with more that 2000 groups of bacteria, becoming difficult to understand this distribution). Anyway, in the MDS that I did (using Bray Curtis similarity), the 2D stress is 0.08. Can I consider that there is no need of transformation?
Indeed, Draftsman plots on sequencing data is difficult to visualise. A 2D stress of 0.08 is good but transformation or normalisation of the data could help to visualise some differences that could be less visible when non-transformed. I usually have a pragmatic approach as the use of a transformation will not be negative, it may not help but it should not be negative in general. So I would try your MDS with and without transformation to see if there is any relevant differences in the ordination.
The normalisation of the data have often a stronger effect that the transformation. Are you expressing the reads number as relative abundance?
Aimeric Blaud I tried different transformations and didnt seem to have big differences between the non transformed. Thank you for the help! Yes, I am expressing the reads number as relative abundance.
- Badges
- Science method