Two additional analyses (which require very different methods are:
Clustering/representation space of disease clusters. This can be done with simple clustering or a representation learning approach.
Collective significant features for a given disease. This is doable with our current tools, may just require repurposing the GPU-based methods to compare individuals, or the p-values comparing distributions. If we are to keep it simple, the Earth Mover's Distance metric may come in handy.
Based off https://github.com/jump-cellpainting/morphmap/issues/2.
Two additional analyses (which require very different methods are: