As a clinician researcher in translational science, I love JMPs ease of use and powerful features, visualization, etc. But there are features that would make it more useful for multi-omics data:
1) support/help/guidance/platforms that focus on small sample sizes with lots of data (i.e. your typical medical basic research where the number of patients is limited but the number of variables per patient is large). A commonly used algorithm that is missing is OPLS-DA, which granted is just an application of PLS-DA which JMP has. You do have lasso which is also popular. Implementing other neural network/machine learning algorithms would be great (XGBoost!). But the point is to integrate the different options into an add-in or platform specific for these tasks, and the guidance/help associated.
2) Support for longitudinal sampling: many times the patients are sampled more than once. Adding that to the above would make JMP stand out and lead the field for these types of analyses.
3) variable selection is crucial, and using different ways is important to see how they perform, i.e. forward addition or backward elimination, bootstrapping (to maintain the #of predictors per variable) vs sub-sampling, etc. Having a profiler for this type of analyses would be amazing, or having a learning curve simulation that shows how the models are expected to perform for given parameters used.
4) The associated visualizations and model comparison and choosing.
5) Another very important feature is network analysis. I don't mean the neural/ML versions, but the ones that use correlations between different variables to create their maps of associations. It's like doing the SEM in the reverse direction. This would be especially useful if done at the individual level i.e. take every patient's scatterplot matrix, and look at the same pairs accross all patients to see which variables are consistently correlated. This is different than putting all the point from all the patients together in one big sample.
Perhaps some of these features exist in separate parts of JMP, and I guess some could be automated/recreated using JSL. But for those of us who are drawn to JMP because it is not R, then having these features organized coherently like other JMP platforms would make a big difference. I can also say that this is the same problem with all the other leading commercial stats packages, so a robust implementation in JMP would be a differentiator.