Quartet Learning

Quartet Learning uses state-of-the art deep learning to infer phylogenetically informative features from almost arbitrary data types.

The input is organized in quartets along a phylogenetic tree. The model produces an output vector for each input, that can be interpreted as the expression level of a designated set of features. The feature vectors of each input quartet are used to calculate a split tree. The loss function then minimizes the edge of the split that is not present in the phylogenetic tree.