Information Technology Reference
In-Depth Information
are less constrained. The side chain at peptide position P1 binds into a deep pocket
while four shallow pockets bind side chains at peptide positions 4, 6, 7, and 9. The
side chains at positions 2, 3, 5, and 8 point toward the T-cell receptor.
We have recently developed an immunoinformatic technique for the prediction of
peptide-MHC affinities, known as the Additive Method, a 2D-QSAR technique
which is based on the Free-Wilson principle (Kubinyi and Kehrhahn 1976), whereby
the presence or absence of groups is correlated with biological activity. For a pep-
tide, the binding affinity is thus represented as the sum of amino acid contributions at
each position. We have extended the classical Free-Wilson model with terms which
account for interactions between amino acid side chains. An Iterative Self-Consistent
(ISC) Partial Least Squares (PLS)-based extension (Doytchinova and Flower 2003)
of the Additive Method (Doytchinova, Blythe, and Flower 2002c; Guan, Doytchinova,
Zygouri, and Flower 2003a) has also been developed for prediction of class II pep-
tide-binding affinity and applied to human class II alleles. We now address binding
to class II human and mouse alleles for peptides of up to 25 amino acids in length.
The ISC additive method assumes that the binding affinity of a large peptide is prin-
cipally derived from the interaction, with an MHC molecule, of a continuous subse-
quence of amino acids within it. The ISC is able to factor out the contribution of
individual amino acids within the subsequence, which is initially identified in an
iterative manner. Using literature data, we have applied the Additive Method to
peptides binding to several human class I (Doytchinova et al. 2002c; Guan et al.
2003a; Hattotuwagama, Guan, Doytchinova, Zygouri, and Flower 2004)
and class II
alleles (Doytchinova and Flower 2003).
Three-dimensional QSARs are a technique of significant value in identifying
correlations between ligand structure and binding affinity. This value is often en-
hanced greatly when analysed in the context of high-resolution ligand-receptor struc-
tures. In such cases, enthalpic changes - van der Waals and electrostatic interactions
- and entropic changes - conformational and solvent-mediated interactions - in
ligand binding can be compared with structural changes in both ligand and macro-
molecule, providing insight into the binding mechanism (Klebe, Abraham, and
ham 1999). Although there are many molecular descrip-
tors that account for free energy changes, 3D-QSAR techniques which use
multivariate statistics to relate molecular descriptors in the space around the ligands,
to binding affinities, have become preeminent because of their robustness and inter-
pretability (Bohm, Sturzebecher, and Klebe 1999). In the case of CoMSIA (Compara-
tive Molecular Similarity Index Analysis), a Gaussian-type functional form is used
so that no arbitrary definition of cutoff threshold is required and interactions can be
calculated at all grid points. The obtained values are evaluated using PLS analysis
(Stahle and Wold 1988).
CoMSIA allows each physicochemical descriptor to be
visualized in 3D using a map, which denotes binding positions that are either
“favored” or “disfavored”.
Recently, CoMSIA has been used to produce predictive models for peptide bind-
ing to human MHCs: HLA-A*0201 (Doytchinova and Flower 2002a) and the HLA-
A2 and HLA-A3 supertypes (Doytchinova and Flower 2002b; Guan, Doytchinova,
and Flower 2003b). We show how CoMSIA has been applied to certain class I MHC
alleles. These models were used both to evaluate physicochemical requirements for
Mietzner 1994; Klebe and Abra
