PathME unsupervised #multiomics
1) genes space → pathways space
2) for each pathway: collapse pathways from multiple omics to one per patient
3) sparse NMF biclustering
✓compared against SNF and iCluster
✓TCGA x 4
✓hyperparameters
✓source code
✓5-fold CV
#SundayMultiOmics
1) genes space → pathways space
2) for each pathway: collapse pathways from multiple omics to one per patient
3) sparse NMF biclustering
✓compared against SNF and iCluster
✓TCGA x 4
✓hyperparameters
✓source code
✓5-fold CV
#SundayMultiOmics
Article: doi.org/10.1186/s12859…
GitHub: github.com/AminaLEM/PathME
Figure © by authors, reused under CC-BY 4.0:
creativecommons.org/licenses/by/4.0
#omics #NeuralNetworks
GitHub: github.com/AminaLEM/PathME
Figure © by authors, reused under CC-BY 4.0:
creativecommons.org/licenses/by/4.0
#omics #NeuralNetworks
Worth noting:
- authors use sNMF consensus from 500 runs (cophenetic correlation + permutation testing to choose # of clusters)
- the autoencoders are denoising
- worth praise is the effort into interpretability (of both features/omics & clinical associations) - see supplement!
- authors use sNMF consensus from 500 runs (cophenetic correlation + permutation testing to choose # of clusters)
- the autoencoders are denoising
- worth praise is the effort into interpretability (of both features/omics & clinical associations) - see supplement!
[Data] Included omics from TCGA: miRNA, mRNA, CNV, methylation data. There is more omics to try (RPPA, MAF - but the number of participants with everything goes down quickly and one needs to deal with missingness). Cancers: LUSC, BRCA, GBM, COADREAD
[Subjective comments] I) "These cancer types were chosen due to the comparable large number of patients with available multi-omics data". This is a good explanation! Not detracting from merits of this work, let's highlight that for an excellent publication one could include more:
e.g. OV, LUAD, KIRC, HNSC (+/or KIPAN, STES); OV has ~550 patients with complete omics of interest (miR + mRNA + CN + methylation), KIRC ~500, which are similar numbers to included GBM and LUSC (see heatmaps from gdac.broadinstitute.org/runs/stddata__…). But having 4 cancers is quite ok!
II) Authors also use golubEsets from Bioconductor as a dataset with ground truth. How good is it?
- golubEsets covariates are "of provenance unknown to the maintainer" (sic)
- only 24% of genes (1729) were actually mapped to pathways. Probably NCI Pathways is not the best idea...
- golubEsets covariates are "of provenance unknown to the maintainer" (sic)
- only 24% of genes (1729) were actually mapped to pathways. Probably NCI Pathways is not the best idea...
III) overall, this seams to be a good quality work, but why is the formatting of the online article so poor in BMC Bioinformatics? Malformed equations, three broken URLs, supplementary word document with a single link inside - like, what is this thing: static-content.springer.com/esm/art%3A10.1…? /n
@threadreaderapp unroll
