Randomly generated model

Recently, Jung et al. [42] developed two artificial neural network models to discriminate intestinal barrier-permeable heptapeptides identified by the peroral phage display experiments from randomly generated heptapeptides. There are two kinds of descriptors one is binary code of amino acid types (each position used 20 bits) and the other, which is called VHSE, is a property descriptor that characterizes the hydrophobic, steric, and electronic properties of 20 coded amino acids. Both types of descriptors produced statistically significant models and the predictive accuracy was about 70%. 

After a random generation of a model population, random mutations of one or a few variables are tried. If better models are obtained after a fixed number of trials, this procedure is repeated on the new generation of models otherwise random mutations of several variables becomes allowed. Also in this case, if better models are obtained after a fixed number of trials, the first procedure is repeated on the new generation of models otherwise systematic addition and elimination of the variables of the population models is performed. If better models are obtained, the procedure restarts from the first step otherwise, if all the variables are checked the procedure ends and the variables of the final population models are checked for their statistical significance and eventually eliminated. 

Only the variables with a c value greater than a cutoff value are retained in the model. The cutoff value of c is estimated in such a way as to exclude from the model all the random variables added to the original variables. The random generated variables range between 0 and 10" °, thus preserving the coefficient variability but negligibly influencing the model. 

---