Interactive demo · Statistical Methods for Modern Biology · Chapter 9

Two tables, twelve samples: RV coefficient and sparse CCA

Microbes (174 OTUs) and metabolites (405 features) measured on the same 12 mice. The RV coefficient compares the two configurations of samples; with so few samples even unrelated tables give RV near √(βXβY)/(n − 1), so it is tested by permutation. Ordinary CCA is useless here: against 3 metabolites, 9 OTUs or more already give a first canonical correlation of exactly 1. Sparse CCA bounds the ℓ1 norms of the weight vectors and selects a few features from each table.

Analysis

Sparse CCA penalties (PMA scale)

OTUs
metab.

Ordinary CCA: number of OTUs p (with 3 metabolites)

Permutations

Scoreboard

Try these

Both tables are log10(1 + x) after the book's filtering; every column is centred and scaled. Penalty c on the PMA scale means ‖u‖1 ≤ c√p. Samples: KO/WT genotype, PD/ST diet (from the sample names).

In the lecture notes