Table 3.

In silico methods to predict T cells epitopes

Method	Description	Reference
Artificial Neural Network (ANN)	Corresponds to a system similar to the brain neural connection, where each cell receives a signal and sends it to another cell. The union between these cells works as a network In an ANN, each cell would be a knot that contains a kind of analysis. One or more entry information are inserted and pass through this knot, resulting, at the end of the network, in different exit information Example: An ANN to predict epitopes creates layers with weights that correspond to characteristics related to the binding affinity between the peptide and the MHC. Thus, by identifying the presence of a certain characteristic, the software goes to the next knot in the network to verify the status of that peptide in relation to another characteristic, and so forth, forming something similar to a status matrix with n characteristics	80,85
NetMHCpan 4.0	Uses an ANN method to predict epitopes using peptide sequences as entry information, and the exit information is generated from the binding affinity data and elution of linkers with mass spectrometer. This method structure is pan because it analyzes just one model, HLA data (Human MHC), and the peptide length	80
Stabilized Matrix Method (SMM)	It is a method that does specificity modeling of sequences of biological processes that can be quantified. When it comes to epitope prediction, it can be used to predict information regarding the peptide capacity to bind to MHC, TAP transport, and proteasome cleavage The entry data corresponds to amino acid or nucleotide sequences, where the coding is done binarily (0 or 1). To each nucleotide sequence, the weight of each residue that can occur in each position of the sequence will be multiplied. The result of this product is the value of prediction y. To measure the efficacy of the process, an experimental average y value will be generated	82–84,86
Support Vectors Machine (SVM)	Through machine learning and statistic learning theory, a model capable of recognizing linear and nonlinear data patterns is created. The data is classified by Kernel functions, linear, radial basis, string, and others For epitope prediction, the SVM is used in the differentiation among peptides that are T cell epitopes from those that are not epitopes	79
NetCTL	It is a prediction method by ANN that uses information about binding affinity, TAP efficacy, and peptide cleavage via proteasomes To measure the binding affinity of each peptide to the MHC-I, values are attributed to each peptide that is inside an interval that has extreme values 0 (low affinity) and 1 (high affinity) To predict cleavage through the proteasome pathway for residues that are used in the NetChop 2.0 C-term 2.0, NetChop C-term, and NetChop 20S-3.0 The TAP transport efficacy is measured through SMM	79,82,83,87–89
NetCTLpan	Epitope prediction in different vertebrate species (pan-specific), amongst which is the human species The NetCTLpan differentiation is the possibility of adjusting different parameters, such as choosing the species; selecting species-specific alleles, and for human studies, it is possible to choose the size of the peptide between 8 and 11-mer; allele selection that is more commonly found in the population; determining the minimal score limit for the prediction and the percentage to consider the prediction as positive (peptides are considered epitopes; defining the proteasome cleavage weights and TAP efficacy, and higher these weights are, higher the possibility of finding epitopes Prediction residues can be seen in two formats, using a graphic that shows the peptides in green as epitopes and in red as non-epitopes, and through a table that shows in columns the MHC prediction values, TAP efficacy, proteasome cleavage score, and the general/combined prediction, and ranking in crescent order of the prediction percentages of a set of peptides with a length of 9 amino acids	79,90
NetChop	Allows the choice of prediction methods named NetChop C-term 3.0 and NetChop 20S-3.0 and allows the alteration of limit score that might interfere with specificity and sensitivity The prediction results can be seen in a similar way to the NetCTLpan, differing only by the table visualization because it presents information related to amino acid residues	79
Consensus	Gather different epitope prediction methods in a single open approach, with the aim of obtaining the best performance of the peptide selection process to those considered epitopes	91