The MemBrain-Contact prediction module is constructed by combining statistical machine learning algorithms and biological residue coevolution analysis from multiple sequence alignments as shown in Fig. 4 [7].
In this paper, the two approaches are compared both as learning algorithms, and as identification modules of an adaptive control system.
First, since the aim of the predictor module is to obtain the best prediction, it was designed to be flexible so that any winner model can be employed among the up-to-date machine learning algorithms.
These types of learning algorithms are called cost-sensitive algorithms.
This allows for efficient learning algorithms.
The driver classification module trains the machine learning algorithm using the feature set fed from the previous module.
The machine learning algorithm module acts as a classifier.
By describing the regulatory network of the core response module in a machine learning algorithm [31], we were able to predict gene expression on a new data set.
Given a cluster of RBPs and their binding sites, the module trains a machine learning algorithm to discriminate between binding sites of two different RBPs, for all possible pairwise combinations of proteins belonging to the same cluster.
The parameters of modular neural networks that are being optimized are: the number of modules (or sub granules), percentage of data for the training phase, learning algorithm, goal error, number of hidden layers and their number of neurons.
