implement Independent Component Analysis (ICA)

[sreichl]

https://hclimente.eu/blog/ica/
Complementary to PCA and (more?) interpretable:

From the website

Application: gene expression analysis
Ultimately, the reason I went down the ICA rabbit hole was not to eavesdrop in parties, but because of its applications in computational biology. Specifically, ICA is used to factorize gene expression matrices into a matrix containing groups of genes that work together (A, the metagenes) and a matrix containing groups of similar samples (S, the metasamples). Both can provide insights into the underlying biology:

The metagenes are particularly useful when we can annotate them with the biological process they capture. We can do that by studying their overlap with known signatures or pathways, or finding commonalities among the genes’ properties.
The metasamples can be helpful if we have other information about the samples (e.g., clinical records) that help us understand what each set of samples have in common.
Teschendorf et al. (2007) provides an example for both. While other matrix factorization algorithms are common, ICA is one of the best, making it an essential tool in the computational biologist’s toolbox.

input: as usual numeric feature table
output: components, but also matrices A for metagenes (meta features) and S metasamples for downstream analysis and interpretation.