TAPREG admits two different types of input queries, a protein sequence or one or several peptide sequences, which are selected by users, prior to entering their sequences. Sequences entered in the server must be in FASTA format. Next follows an example of these two types of input.

       The two models available in TAPREG are based on support vector machine regression and were trained on sparse (DS_613 sparse model) and blosum (DS_613 blosum model) representations of 613 9mer peptides of known binding affinity to TAP. Models were only trained in peptide residues 1, 2, 3 and 9, which have been shown to have the major contribution to TAP binding The binding binding affinity of any peptide randing from 8 to 16 is computed by applying the models to peptide fragments encompassing the noted residue selections. The server will return an error for input peptides out of that range.

      In cross-validation, the model trained on blosum-encoded sequences achieved a person correlation coefficient (Rp) of 0.83 ± 0.03 while the sparse counterpart achieved an Rp = 0.81 ± 0.03. The correlation achieved by the DS_613 blosum and sparse models predicting the known TAP binding affinity of a set 98 peptides with a length ranging from 10 to 16 residues (independent test set) was 0.78 and 0.73, respectively.

• Protein
The server will compute the TAP-binding affinity (IC50) of each 9mer peptide encompassed by the protein and return them sorted out by their binding affinity in a Table. The lower the IC50 the larger the affinity. If a threshold was provided to the server, only those peptides with a binding affinity above that threshold will be returned. Here follows a representative output.

• Peptide(s)
The server will compute the TAP-binding affinity of the peptides and will sort them out according to their binding affinity, as shown in the example bellow.