Integrate DEGAS_contributionVar with DEGAS_contributionGene

[bschilder]

If you run an integrative algorithm on these two sets of reductions, you could get some useful variant-gene relationships (both cis- and trans-).

MOFA+ seems like a good tool for this, since you're essentially learning multi-modal relationships from the same set of samples. MOFA+ also induces sparsity constraints and can even model variation due to hierarchical metadata attributes. Plus you can accelerate with GPUs!