research considers all potential actors in an immune response, the
pharmaceutical industry has traditionally considered the ability of a
vaccine preparation to induce complement-mediated in vitro killing
of the targeted bacteria to be an essential condition, as this tends to
correlate with vaccine efficacy in humans [5, 74]. Subcellular localization can be predicted, for example, with PsortB [50], which
stands as one of the most reliable applications for this matter.
3.3.3 Protein Solubility
Apart from the annotated data it is important to have information
on structural features and physical-chemical properties. Selection of
a protein that is not soluble or that will give problems in the
production and purification process should be avoided. Likewise,
selection of an epitope that is highly hydrophobic or has a propensity to aggregate for peptide production should be also avoided.
These issues can be addressed by prediction of transmembrane
helices [51], aggregation propensity [53], surface accessibility [54],
and unstructured regions [52]. The information provided by these
analyses can be very valuable for the prioritization of predicted
antigens and epitopes.
3.4 Epitope
Prediction
In general, we identify a protein as a potential antigen through the
prediction of epitopes. Currently, we may predict protein fragments
binding to MHC class I or MHC class II molecules as well as
protein regions physicochemically amenable to the binding of antibodies. The accuracy of predictors may in some cases exceed 80%
(particularly for MHC class I), but one should rather count on an
average performance of 70%, and even that may be very optimistic
depending on the alleles involved. Peptide processing predictions
may be used to enhance the accuracy of MHC class I binding
predictions [75], although when dealing with bacterial pathogens
one will be more often interested in MHC class II binding, which
predictions are per se less reliable. The prediction of TCR recognition is in its infancy and no reliable prediction tools exist yet.
3.4.1 MHC Binding
Affinity Analysis
There exist several methods and programs for this purpose. The
IEDB database offers a tool implementing several of these methods
for both MHC class I and II peptide binding prediction [55]. We
recommend this tool, as it ranks amongst the best in terms of
accuracy and provides a common interface for class I and II input
and output, allowing the test of different methods using the same
input data and reading the results in the same format. If a local run
is required, we usually use standalone programs implementing
NetMHCpan methods for both MHC class I and II [56, 57]. For
MHC class I, the software NetMHCcons implements multiple
prediction methods and presents a consensus as a result [58]. However it is slower, more complex to install and uses an older version
of the NetMHCpan method.
52
Daniel Yero et al.
pharmaceutical industry has traditionally considered the ability of a
vaccine preparation to induce complement-mediated in vitro killing
of the targeted bacteria to be an essential condition, as this tends to
correlate with vaccine efficacy in humans [5, 74]. Subcellular localization can be predicted, for example, with PsortB [50], which
stands as one of the most reliable applications for this matter.
3.3.3 Protein Solubility
Apart from the annotated data it is important to have information
on structural features and physical-chemical properties. Selection of
a protein that is not soluble or that will give problems in the
production and purification process should be avoided. Likewise,
selection of an epitope that is highly hydrophobic or has a propensity to aggregate for peptide production should be also avoided.
These issues can be addressed by prediction of transmembrane
helices [51], aggregation propensity [53], surface accessibility [54],
and unstructured regions [52]. The information provided by these
analyses can be very valuable for the prioritization of predicted
antigens and epitopes.
3.4 Epitope
Prediction
In general, we identify a protein as a potential antigen through the
prediction of epitopes. Currently, we may predict protein fragments
binding to MHC class I or MHC class II molecules as well as
protein regions physicochemically amenable to the binding of antibodies. The accuracy of predictors may in some cases exceed 80%
(particularly for MHC class I), but one should rather count on an
average performance of 70%, and even that may be very optimistic
depending on the alleles involved. Peptide processing predictions
may be used to enhance the accuracy of MHC class I binding
predictions [75], although when dealing with bacterial pathogens
one will be more often interested in MHC class II binding, which
predictions are per se less reliable. The prediction of TCR recognition is in its infancy and no reliable prediction tools exist yet.
3.4.1 MHC Binding
Affinity Analysis
There exist several methods and programs for this purpose. The
IEDB database offers a tool implementing several of these methods
for both MHC class I and II peptide binding prediction [55]. We
recommend this tool, as it ranks amongst the best in terms of
accuracy and provides a common interface for class I and II input
and output, allowing the test of different methods using the same
input data and reading the results in the same format. If a local run
is required, we usually use standalone programs implementing
NetMHCpan methods for both MHC class I and II [56, 57]. For
MHC class I, the software NetMHCcons implements multiple
prediction methods and presents a consensus as a result [58]. However it is slower, more complex to install and uses an older version
of the NetMHCpan method.
52
Daniel Yero et al.