Models’ performance on IC80 data.

The area under the precision-recall curve (PR AUC) and the binary cross-entropy (Log Loss) are reported. Models were trained on sequences paired with phenotypes defined using IC80. GBM is Gradient Boosting Machines; RF is Random Forests; LBUM is language-based universal model; ENS is the ensemble model that averages predictions from GBM, RF and LBUM. The red shade means that LBUM had a better score than both RF and GBM models. The blue shade means the ensemble model scored better than all three individual models. Numbers between parentheses are standard deviations. (XLSX)