methClust is a software that fits Grade of Membership (GoM) model, much like CountClust. The difference between methClust and CountClust lies in the model assumptions. CountClust assumes a Multinomial model for the counts for each sample and hence bases the clustering on the compositional content of the samples. methClust on the other hand assumes a Binomial model for the counts for each sample and ach feature and is applicable when there is a constraint on the size of the counts per cell but the total counts for a sample is not fixed.
A detailed report on the broad results from the analysis are presented here.
For detailed breakdown of the analysis along with the relevant codes are presented below.
We first apply a special elementary version methClust on Bernoulli or presence absence data. The immediate application it targets is ecological presence absence data of species in different sites.
GoM model for binary presence-absence data : Model framework
Simulation test of binary version of methClust with presence-absence data
methClust on binary presence-absence taxonomic data: White et al (2018) Birds
methClust on binary regional assemblage data: White et al (2018) Birds
We applied Binary methClust to the global presence absence data, as demonstrated here and here. This analysis fitted the model on the Cells by Birds data matrix. Alternatively, we can also fit the model on the Birds by Cells data matrix, as depicted here.
For the geoStructure representation examples, as in the example, check out the R code here.
We also applied Binary methClust to presence absence data for individual continents - Australia, Europe, Africa, Indian subcontinent, South America, North America. We found the GoM Structure plots to closely follow the climate patterns.
For the geoStructure representation examples, as in the example, check out the R code here.
Also, when restricted to the presence absence data of bird species , along the Wallace Line, Binary methclust detects Wallace Line as a transition boundary between two bird ecozones, check our analysis here.
The top bird species that drive the clusters - Wallacea, Australia.
We now look at the model framework and applications of a more general form of methClust.
Simulation example application of methClust
Application of methClust on Seb’s hCh data - preliminary run
For the geoStructure representation, as in the example here, check out the R code here
For any queries related to the methclust package, contact Kushal K. Dey here kkdey@uchicago.edu
This R Markdown site was created with workflowr
