Bray Curtis for functional dissimilarity

Hi,

I need to know how to run a Bray Curtis calculation for functional dissimilarity, including which program to run to actually produce a plot. I'm presuming something like SYN-TAX 2000 ordination can do it, but I'm not sure how to input the variables.

I basically have 100 individuals of two similar species at two different sites; so 50 individuals of one species at one site and 50 individuals of another species at another site. I have measured key morphological features for each individual, totalling 10 measures for each. There are slight morphological changes between species, some morphological traits are more similar than others.

How do I create a visual dissimilarity plot using the data? I'd like something like a scattergram with 2 axes and ideally the scattered plots for each species will separate and group, thus showing the morphological differences between the species. It would be ideal if the output plots were also numbered for each individual to confirm which site they come from.

Thank you,

Statistics Novice.

dissimilarity bray curtis

• 2,731 views • link

updated 5.2 years ago by Kevin Blighe

• written 5.2 years ago by statisticsnovice •

1 answer

Why do you want to use the Bray Curtis dissimilarity? It's meant to compare the species composition between two populations but here you seem to want a measure of (dis)similarity between feature vectors. For this pick any distance measure that capture a relevant notion of similarity. You can get a 2d plot using a dimensionality reduction method. Without resorting to computing a distance matrix, you can try PCA and plot individuals on the first two components. PCA has the advantage that the components are easily interpretable in terms of the original features. Or if you want to start with a distance matrix, use multi-dimensional scaling to plot the individuals on the first two dimensions. Or use t-SNE or UMAP.
All of this can easily be done in R.

• 0 views • link

written 5.2 years ago by Jean-Karim Heriche

Thanks for the response.

I suppose the answer is that I'm a statistics novice and really need some assistance. Think of it like this, my destination is the plot, the method is the highway and I get there by learning to drive. I haven't learnt to drive yet, so telling me to use PCA, t-SNE or UMAP in R is like telling me which highway I have to take, when I don't even know how to start the car! I think though that multi-dimensional scaling is probably the way to go, but why just use the first two dimensions? I want to compare all the dimensions I have.

• 0 views • link

written 5.2 years ago by statisticsnovice •

Please use the appropriate "Add reply" button to reply to a comment. Creating an answer makes the topic appear as having been answered.
Dimensionality reduction methods combine information from all the original features/dimensions into typically a smaller number of dimensions. One typically start with plotting the first few dimensions as they are expected to capture more information than subsequent dimensions. I suggest you do some reading on dimensionality reduction methods, starting with PCA to at least get some understanding of what these methods do. It may also be worth looking for some knowledgeable person near you who can provide more directed and timely guidance than what's possible on a forum.

• 0 views • link

written 5.2 years ago by Jean-Karim Heriche

Hey mate, I was able to move the posts around to align them better.

• 0 views • link

written 5.2 years ago by Kevin Blighe