Information Technology Reference
In-Depth Information
peptides without using canonical anchors, with extra affinity arising from other in-
teractions made by the rest of the peptide. This is also likely to be a feature of class II
binding. For example, Liu, Dai, Crawford, Fruge, Marrack, and Kappler (2002)
showed that for I-A
b
it was possible for a peptide bearing alanines to bind to its four
main pockets - which correspond to positions P1, P4, P6, and P9 and which usually
bind larger peptide side chains - with compensatory interactions made by residues at
other positions in order to maintain overall affinity. Our class II models suggest that
the relative contributions, of particular residues, to binding are spread more evenly
through the peptide than is generally supposed, rather than being concentrated solely
in so-called anchor positions.
The ISC algorithm described above combines an iterative approach to selecting
the best predicted binders with PLS, a robust multivariate statistical tool for model
generation. The ISC method is universal in that it can be used for any peptide-protein
binding interaction where the peptide length is unrestricted but the binding is limited
to a fixed, if unknown, part of the peptide. Implementation of the method is straight-
forward, it is fast to use, and its interpretation is straightforward. The final models
derived from these calculations will be included in an updated version of MHCPred
(Doytchinova and Flower 2003; Guan et al. 2003a; Hattotuwagama et al. 2004;
Guan, Doytchinova, and Flower 2003c; Guan, Doytchinova, Zygouri, and Flower
2003d).
4.5 Conclusions
From our studies, we find that distinct MHC alleles, both class I and class II, exhibit
different peptide specificities: peptides are bound with particular sequence patterns,
leading to the development of so-called motifs (Takamiya, Schönbach, Nokihara,
Yamaguchi, Ferrone, Kano, Egawa, and Takiguchi 1994). Motifs are usually expressed
in terms of anchor residues: the presence of certain amino acids at particular positions
that are thought to be essential for binding. Taking human class I allele HLA-B*3501
as our example, previous studies have indicated the need for anchor residues at posi-
tions 2 (Pro) and 9 (hydrophobic or aromatic residues, such as Phe, Met, Leu, Ile, and
especially Tyr). Primary anchor residues, although generally deemed to be necessary,
are not sufficient for peptide binding, and secondary anchors, residues that are favor-
able, but not essential, for binding, may also be required; other positions show posi-
tional preferences for particular amino acids. Moreover, the presence of certain resi-
dues at specific positions of a peptide can have a negative effect on binding (Amaro,
Houbiers, Drijfhout, Brandt, Schipper, Bavinck, Melief, and Kast 1995; Sidney, del
Guercio, Southwood, Engelhard, Appella, Rammensee, Falk, Rötzschke, Takiguchi,
and Kubo 1995; Smith et al. 1996b). Although motif methods are admirably simple -
easy to implement either by eye or more systematically scanning protein sequences
computationally - there remain many problems with the motif approach.
Although it is possible to score the relative contributions of primary and secondary
anchors to produce a rough-and-ready measure of binding affinity (Amaro et al. 1995;
Sette, Vitiello, Reherman, Fowler, Nayersina, Kast, Melief, Oseroff, Yuan, and Ruppert
1994b), the most significant problem with the motif approach is that it is, fundamentally,
