We then built a gradient boosting classifier that takes these feature sets as input. One classifier takes the 40 features per protein fragment as input, and produces a probability for the secretability of that fragment. After training a classifier for each property, an ensemble model is constructe...