GTEx experiments

[millerh1]

Run experiments on GTEx

• [x] GTEx classification by tissue type (one vs rest)
• [x] Compare to NMF and PCA on average precision and / or F1
• [x] Sweep params and look at the performance
• [ ] Difference from least squares solution of multiPLIER. We would have to think about this. (low priority)
• [x] UMAP plot of the samples clustered together
• [x] Take a one-vs-rest classifier (brain or heart), look at the top LVs, and examine the pathways, are they interpretable?
• [x] Try using cell-type specific pathway databases, as in those derived from GTEx (see enrichr pathways here) -- this is a sanity check for interpretability
• [x] Try using Hallmark pathway database
• [x] Try C2 Pathway database
• [ ] #53

[millerh1]

Average precision or F1 on the Y axis

[millerh1]

For each one -- change the sparsity (0.3, 0.5, 0.7, 0.9)

Shows relationship between sparsity an performance across these different databases.