Biology Reference
In-Depth Information
Towards this aim, in addition to the programs already mentioned,
other systems are available to researchers. The MULTIPRED pro-
gram
216
(available at http://antigen.i2r.a-star.edu.sg/multipred/) is
based on ANN methods to predict HCV T-cell epitopes for MHC
class I A3 molecules and human papillomavirus (HPV) type 16 T-cell
epitopes for MHC class A2 variants, and is reported to achieve good
sensitivity and specificity and prediction capability, with an area under
the receiver operating system curve (A
ROC
) >0.80. NetChop server
215
(available at http://www.cbs.dtu.dk/services/NetChop/) uses ANN
method to predict proteasome cleavage sites responsible for generat-
ing CTL epitopes
ANN-based predictions can also reduce the number of bench
experiments needed for T-cell epitope screening. ANN applications
allow investigators to reduce and focus the number of candidates for
identification of T-cell epitopes while increasing the efficiency of the
search space for these candidates. For example, Honeyman and col-
laborators (1998) synthesized 68 peptides (9-mers) and used a
trained ANN to perform binding prediction on this set. After com-
pleting experimental binding and T-cell response tests, they show that
ANN-predicted binding would have reduced the number of synthe-
sized peptides from 68 to 26, with the potential loss of four epitopes.
In another case, Schönbach and collaborators (2002) applied ANN
methods to the large-scale identification of HIV T-cell epitopes from
Pol, Gag, and Env sequences for HLA-A*0201 and found 890 HIV-
1 and 232 HIV-2 epitope candidates. The overall sequence coverage
of the predicted A*0201 T-cell epitope candidates was 2.7% for HIV-
1 and 3.0% for HIV-2. By comparing the ANN-predicted binding
with other bioinformatics methods and experimentally confirmed
A*0201-restricted epitopes, they were able to extrapolate their results
and estimate that approximately 247 predicted HIV-1 are yet to be
discovered as active A*0201-restricted epitopes. Finally, with proper
encoding, ANNs can also be used for feature extraction. Conclusions
can be drawn regarding the relative importance to binding of specific
positions in the peptide and chemical properties of amino acids at
those positions. Thus, use of these capabilities can facilitate the
de novo
-
driven design of viral vaccine or immunotherapeutic candidates.
