Predicting the network of substrate-enzyme-product triads by combining compound similarity and functional domain composition