Epitope Prediction

Epitope is the antibody-binding region on the cognate antigen. Predicting what kind of antigen surfaces bind to a known antibody is useful in reverse engineering the antibody to target specific antigen.

EpiPred is an algorithm for such prediction based on graph models. The score imposes greater weights on clusters of interactions, ie. the more robust parts of the network. To validate this approach, native and decoy (generated from global docking) structures are compared and ranked.

Syntax for Scoring Function:

• $G =$ graph
• $Epi =$ residues in proposed epitope patch
• $Ab =$ residues in the binding site on the antibody
• $n =$ node – Cartesian product of $Epi$ and $Ab$: $Epi \times Ab$:
• Eg. $n_1 = r_{ab1} \times r_{ag1}$$n_2 = r_{ab2} \times r_{ag2}$
• an edge exists between $n_1$ and $n_2$ if $|dist(r_{ab1},r_{ab2}) -dist(r_{ab1},r_{ab2})|<1\AA$
• $d(n) =$ the degree of node $n$
• $latex Pr(T_{ab},T_{ag}) =$ likelihood of the docking algorithm to correctly pair residues $T$ on $ab$ and $ag$.

Then the score of a putative epitope $Epi$:

$EpitopeScore(Epi,Ab) = \sum_{n \in G} d(n) Pr(T_{ab},T_{ag})$

Evaluating Global Docking and rescoring:

$DecoyScore(d) = \sum_{r_{ab} \in Ab; r_{ag} \in Epi; dist(r_{ab},r_{ag})<4.5 \AA} Pr(T_{ab},T_{ag})$

To assess the quality of decoys, interfacial root mean square deviation ($Irmsd$) is calculated between the interface region of the native and decoy structures superimposed on each other. $Irmsd < 10 \AA$ is considered close-to-native.

Source:

Krawczyk et al., 2014