Still a bit of a mess

[bethydavis]

I think I'm still a bit confused, or more likely overthinking things

What scenarios/what data would I need in order to use the MESS model? What types of questions would I use the MESS model for? For machine learning in this model training context, are there "standard" or general agreed upon things for what your corpus should have or should avoid for training?

[phoebejek]

I have the same first few questions as Beth...thanks for asking those, Beth!

Also, why will (environmental) filtering communities have less variation in trait values than neutral and those assembled by competition would have more variation?? Can we discuss this a bit...

Why do you think the non-neutral models have less classification accuracy?

[ajrominger]

Super great questions! Here are some answers:

• for the types of questions: any question you would ask with neutral theory or competitive theory you could ask with MESS, part of the idea is that MESS should just do a better job with these same questions because it can utilize more data types. But you can also ask new questions like how do different processes (competition, past perturbations, etc) impact not only species richness and abundance but also genetic diversity, traits, evolutionary relationships
• for the types of data: MESS is designed to handle missing data, so you could present it with just species abundances data, or just sequence data (e.g. metabarcoding), but it does best when you present it with abundances, genetics, and traits (e.g. body size)
• for the machine learning aspects: this is still really new, so there are not "standards" yet. We've done enough experimenting to know that the Hill numbers are very useful, and other published works point to Hill numbers being useful (look back to Renata's lecture on that)
• as to why filtering has less trait variation this is because (in MESS and likely/hopefully in the real world too) only a small number of traits (or a small range of trait values) are well-suited to the environment that is imposing the filter. So almost by definition there has to be less trait variation in the environmental filtering case
• for by non-neutral models have less classification accuracy this is likely because both filtering and competition represent forces that act to limit the types of traits that are advantageous in the local community---they are different types of limitations, but still limitations, especially compared to neutral, which has no trait-specific limitations. Thus the patterns these two processes produce in the data end up being more similar to each other than the patterns produced by neutral models. If the patterns are more similar, then the classifier will have a harder time classifying them