Experiments were carried out as follows: the facial datasets built in the experimental phase 1 was used to set the projection (through LDA), classification/regression (through SVM), and scaling models.
A schematic representation of the dataset built up is shown in Additional file 1: Figure S1.
The variables elected for the present analysis were obtained in a dataset built from 1999 to 2007.
In this paper, based on the dataset built by Liu et al. [ 27] in our group, amino acid composition, motif frequency, predicted secondary structure information, and the interaction of predicted secondary structure segments were applied for the recognition of protein folds.
These three datasets were built in order to tackle different questions.
Therefore, a concatenated alignment of the HK and RR datasets was built in order to increase the phylogenetic signal.
Since the VM training and testing datasets were created by randomly assigning 4/5 of individuals for training and 1/5 for testing, we combined them for the purposes of creating a testing dataset for models built in the ARMA dataset.
We used GeneSpider and a linear model to generate datasets, while GNW has nonlinearities built in to its dataset generation.
Also, we show the relationship between both metrics in three real datasets built from digital libraries of distinct fields Computer Science with DBLP2, Medicine with PubMed3, and Physics with APS4 ("Neighborhood overlap and absolute frequency of interaction" section).
A Neighbor-joining (NJ) tree [ 78], shown to be a useful clustering method for large datasets [ 79, 80], was built in MEGA for the initial data set using the following parameters of BOLD: Kimura 2-Parameter (K2P) distance model [ 81] with pairwise deletion of gaps/missing data and inclusion of all substitutions (transitions and transversions).
