Benchmark of multi-omics joint Dimensionality Reduction (jDR) approaches in CMI-PB

I have taken the original momix-notebook repository and investigated 5 representative jDR approaches.

The benchmarked methods are:

• Integrative NMF (intNMF)
• Joint and individual variation explained (JIVE)
• Multiple co-inertia analysis (MCIA)
• Regularized Generalized Canonical Correlation Analysis (RGCCA)
• matrix-tri-factorization (scikit-fusion)

Due to installation or running difficulties the following methods have been skipped (as of 2021.08.09):

• iCluster
• Multi-Omics Factor Analysis (MOFA)
• Multi-Study Factor Analysis (MSFA)
• tensorial Independent Component Analysis (tICA)

As part of this analysis I take a look at:

1. ability to identify factors associated with task data
2. metagenes associated with biological annotations (Reactome, GO, Hallmarks)

Input data

The input data is derived from the CMI-PB data tables.

Cite momix

The preprint describing momix is available in BioRxiv https://www.biorxiv.org/content/10.1101/2020.01.14.905760v1