search

gwaybio / pancan_viz

Visualizing TCGA pancancer datasets
BSD 3-Clause "New" or "Revised" License
1 stars 1 forks source link

Is PCA coming soon? #7

Closed rdvelazquez closed 7 years ago

rdvelazquez commented 7 years ago

Nice work @gwaygenomics! Is there any plan to include dimensionality reduction via PCA as was done here?

gwaybio commented 7 years ago

Yeah definitely! See #2 - ideally, I'd like to slowly add new features improved functionality as well

rdvelazquez commented 7 years ago

Gotcha. I'll stay tuned for PCA. That will be interesting to see and mess around with as it relates to the current work being done with cognoma.

Another interesting variable to plot might be how mutated the sample is. You could do total number of mutations, log(mutations) or bin the total number of mutations. I'd offer to help out but I'm not very good with R (yet).

It's interesting how undifferentiated the diseases/organs seem to be with this dimensionality reduction technique compared to PCA. What method of dimensionality reduction is this using?

gwaybio commented 7 years ago

Another interesting variable to plot might be how mutated the sample is. You could do total number of mutations, log(mutations) or bin the total number of mutations. I'd offer to help out but I'm not very good with R (yet).

I agree - mutation burden is a large signal. It would be interesting to visualize here

It's interesting how undifferentiated the diseases/organs seem to be with this dimensionality reduction technique compared to PCA. What method of dimensionality reduction is this using?

It's a variational autoencoder. I suspect that part of the reason for the lack of separation is that the algorithm is trained sub-optimally. This was a first pass, I am still tuning some parameters

gwaybio commented 7 years ago

@rdvelazquez checkout updates https://gregway.shinyapps.io/pancan_plotter/

Added functionality of data type / algorithm selection. Still WIP so any suggestions/contributions are welcome!

Closing this issue since PCA is here (cf8b29913646d2b85c1d1e6dcbe55286bfcedf0c)