Analyzing coral health dataThis is more of a conceptual question than one that could be solved by writing a script (though please correct me if I am wrong!), and I am not even sure if it is appropriate for the JMP community, but here goes nothing:
I study coral reefs, more specifically the reef corals upon which these ecosystems are based. As many of you probably already know, coral reefs around the globe are threatened by an onslaught of anthropogenic stressors, most notably climate change; indeed, no reefs have been spared from the wide-ranging impacts of humankind (i.e., there are, sadly, no pristine reefs left). Consequently, there has been an urgent push to 1) better understand the fundamental biology of key coral reef inhabitants (most notably the reef-building corals themselves), 2) understand how these organisms will respond to future changes in their environments, and 3) make data-driven predictions as to which species/populations are most stress sensitive. Despite coral reef research growing at an ever increasing rate, there are inherent issues with the interpretation of coral reef health data that have prevented us from addressing the aforementioned questions. Two such issues include:
Issue #1: There are no healthy, control reefs against which to compare results of, for instance, biomarker surveys. Do we even know what constitutes a healthy coral in the absence of data obtained from the pre-Industrial era?
Issue #2: Earth’s oceans are in a constant state of flux; does it make sense, then, to use “static” statistical tools (e.g., ANOVA) to analyze coral reef data, or should we instead be using something more dynamic like predictive modeling?
Despite these two major issues, I have collected coral physiology data from reefs across the globe in the hopes that I can ultimately settle on a statistical pipeline that will allow me to assess coral health. Please keep in mind that we cannot simply carry out tank experiments, experimentally stress the corals, and then use this to drive our field studies (before, I had been doing tank studies for just this reason). Yes, I know people DO do this, but it is actually invalid. The reasons are because a "control" coral used in the hypothetical study is likely to be stressed already (due to climate change). Secondly, a "control" coral in a highly impacted place like Taiwan (where I work) may constitute a "stressed" coral in a relatively more pristine location. Also, common "stress" markers (e.g, HSPs) studied in model systems (e.g., humans) are expressed at high levels at all times in corals (or else I could use a biomarker approach).
I have been using JMP to 1) model coral behavior in an endosymbiosis-tailored, statistically rigorous framework and 2) make predictions as to which colonies/reefs may be at most risk. Specifically, I am interested in the idea that VARIABILITY in coral response (as in, the variation across response variables within a single sample), rather than the mean response of individual response variables measured, is a more useful metric for those interested in forecasting behavioral change in reef corals and other marine invertebrates. This is based on what we know about transcriptional "chaos" and cancer. Maybe, though, I have become myopic from having been focused on this issue for so long and am missing something simple! If anyone has any ideas on how to glean the most insight from my coral physiology datasets (one of which is cited below and posted on JMP's website), I would be happy to hear from you. Thank you in advance for your time, and apologies for the long post.
Here is the aforementioned sample dataset I have recently posted for those interested in seeing what the data look like: https://community.jmp.com/t5/JMP-Sample-Data/Reef-coral-data-from-the-Austral-Islands-French-Polynesia-and/ta-p/61858
Re: Analyzing coral health dataFunny that you mention PLS because not only did I attend your excellent PLS webinar a few weeks ago, but I also won a PLS book from JMP (random drawing of names) during said webinar (thanks to Gail Massari)! You may recall that I asked a question about why all factors are retained in the model, even when the goal is to simplify it (I now know why and have been "pruning" my own models to see if I can simplify them without losing predictive power.). I am halfway through the book, and it has given me some great ideas about analyzing my coral transcriptome data, in which your estrogen receptor example is basically exactly what I want to do: make predictions about coral physiology based on a simpler gene expression model. Now the issue is not that I lack JMP Pro but that a PLS with 120,000 columns (gene expression data) is causing JMP to crash (I only have 8 GB of RAM)!
Re: Analyzing coral health dataYou should be able to find a very rich literature set of case studies, examples, papers, etc. in the bioinformatics space for humans. Quite frankly, from my seat as a JMP systems engineer, the field of what is often called 'translational research' in the bio tech and pharmaceutical industry are the Holy Grail of much of what Dr. Shmeuli covers.
In fact for one small, very small example of this sort of thing take a look at the Advanced Mastering JMP event I host where the subject is an analytics technique called partial least squares. The last case study is one where I show how human genetic expression data can be used to build a classification model for predicting estrogen receptor status in humans. Unfortunately you won't be able to duplicate what I show without JMP Pro. 
https://www.jmp.com/en_us/events/ondemand/mastering-jmp/partial-least-squares.html
Re: Analyzing coral health dataThanks for your input. I only have regular JMP (not JMP Pro), but I will check out this video. I am interested in making the transition from explaining what has already happened (what I have done statistically for the past 15 years) to predicting what will happen next (using the data acquired over my career to date), s othis video seems right up my alley.
I am thinking about applying market analytics and that sort of thing to making predictions about coral behavior and health. It seems like JMP has all kinds of cool tools for market analysts, as well as engineers testing new products, that we biologists never touch.....but should! I feel like someone should have worked this out in human medicine (e.g., making predictions of one's health based off of molecular biomarkers), so perhaps I won't need to reinvent the wheel, if I dig into the literature deeply enough.
I have never heard of variance components modeling (VCM), but I will look into it, as it appears to be a feature in (regular) JMP 14. In the past I have used JMP's predictor screening function to determine which response variables measured contribute most to aberrant behavior in my samples (essentially those which scale positively with the Mahalanobis distance). It seems to be a similar idea to VCM: identifying the major contributors to variance observed (right?). Stepwise regression (another "new" approach for me) has also helped immsensely in narrowing down the large number of potential environmental factors of interest (e.g., temperature, depth, salinity, light, etc.) in terms of giving me a best-fit model with a relatively low BIC. In fact, I was so excited by my first attempt at stepwise regression (w/ JMP13) that I just wrote a whole paper about applying it to coral reefs!
In the event of breakthroughs (e.g., from taking the approaches you have outlined in your response) or statistical epiphanies, I will report back to this post such that other environmental physiologists who find themselves asking similar questions about how to advance our understanding of our rapidly changing world may potentially benefit.
Re: Analyzing coral health dataTwo thoughts for you...one is macro the other micro...
Macro: Have you taken a look at Galit Shmeuli's work on Expaining vs. Predicting? Here is a link to a JMP Analytically speaking event where Dr. Shmeuli articulates from a modeling and analytics perspective important issues that influence the 'how' one should first collect, and then analyze data with an emphasis on empirical modeling.
https://www.jmp.com/en_us/events/ondemand/analytically-speaking/to-explain-or-predict-that-is-the-question.html
Micro: Within JMP Pro there are some modeling techniques that might be of value in the Generalized Regression platform with modeling personalities for quantile regression where modeling quantiles is more important than 'means'. Or what about variance components modeling?
