Figure 3.

Grouped, nested fourfold cross-validation scheme to train and compare selected machine learning models. RBP sequences were grouped based on computed identity percentages between every sequence pair. In the inner loop, every model’s hyperparameters were optimized using grid search (with groups as defined before). In the outer loop, the performance of every optimized model was measured. In the outer loop as well, the defined groups control which sequences are in which fold.