Information Technology Reference
In-Depth Information
4.3.3 Comparative Molecular Similarity Index Analysis (CoMSIA)
For each of the 12 alleles, all peptides were built and aligned in three dimensions
(Fig. 2), their geometry was optimized, and AM1 (Dewar et al. 1985) calculations
performed within SYBYL 6.9. The peptides were placed within individual 3D grids
(Fig. 3). The final settings for the three models are shown in Table 6. The generated
models (
n
=30-236) have an acceptable level of predictivity: LOO-CV statistical
terms, SEP and
q
2
, ranged between 0.443 and 0.889 and 0.385 and 0.700, respec-
tively. The non-cross-validated statistical terms NC, SEE, and
r
2
ranged between 4
and 12, 0.071 and 0.411, and 0.867 and 0.991, respectively.
To generate CoMSIA coefficient contour maps for each allele, which describes
the relationship between the binding affinity and each physicochemical descriptor,
three non-cross-validated “all fields” models were created based on the five phys-
icochemical descriptors (steric, electrostatic, hydrophobic, hydrogen bond donor
and acceptor). The descriptors involved in the interaction between the peptide and
the MHC molecules are presented in the coefficient contour maps as shown in
Fig. 4 for the H2-D
b
allele. For simplicity, the interaction between only one peptide
and its respective contour map is shown with the N-terminus to the left and the
C-terminus to the right. Table 7 shows a summary of the position specificities
between the physicochemical descriptors and peptide positions for the A2 super-
motif and class I mouse alleles.
b
Fig. 2.
Superimposed alignment of peptide molecules for the H2-D
allele.
