2.6 | Protein-ligand complexes
Accurate computation of binding modes for organic ligands is of great
practical relevance because of its role in rational drug design. There
have been several previous challenge experiments in this area, most
notably the D3R series (39), but all have ceased operation. Ideal
targets are likely those related to drug development, but in this first
CASP round, only a small number of endogenous ligands were available.
Additionally, participants were not provided with structures for the
host protein (and one RNA) structures so that if a successful model of
the host macromolecule was not generated, ligand docking was doomed to
failure. Thus, this should be regarded as a pilot experiment and not a
definitive assessment of the state of the art. Nevertheless, some useful
insights were obtained by the assessor (14). First, participating deep
learning methods were not as effective as traditional ones. Second, of
the traditional methods, those that worked from homology or analogy
rather than de novo docking were most successful. These are methods in
which the PDB is searched for similar ligands and binding pockets,
providing a starting point for pose refinement. Perhaps not
surprisingly, less flexible ligands tended to more tractable.