(a) (b)

a) The RVM model employing the Gaussian kernel function for a two-cluster

he triangles and filled dots stand for two clusters and the large circles stand for

ectors found by RVM. (b) The ROC curve of a Dayhoff kernel bio-relevance

hine constructed for the factor Xa protease cleavage data. The AUC was 0.962.

ep neural network

eural network (DNN) is an algorithm of deep learning

huber, 2015; Bengio, et al., 2015]. The basic principle of DNN is

ate feature extraction into a model construction process by

g the number of hidden layers, where the first hidden layer is used

ature extraction and selection process [Bengio, et al., 2015]. DNN

n used for biological pattern discovery such as human

ylation coding mechanism [Leung, et al., 2018] and signal

ecognition [Savojardo, et al., 2018]. One of the available R

for DNN is H2O.

he factor Xa protease cleavage data, a grid search was used to

r the optimal hidden neurons for two hidden layers of the H2O

he optimisation was varying the hidden neuron number in both

yers from two to 20. The feature space was generated using the

matrix to measure the similarity between peptides. Therefore this

as called a bio-DNN model. The grid search led to nine hidden

for the first hidden layer and five neurons for the second hidden

he optimal DNN model structure. The first step of using the H2O

was to initialise a H2O environment using the following code,

init(nthreads=-1,enable_assertion=FALSE)