A similarity-driven approach to flexible ligand docking is presented. Given
a reference ligand or a pharmacophore positioned in the protein active sit
e, the method allows inclusion of a similarity term during docking. Two dif
ferent algorithms have been implemented, namely, a similarity penalized doc
king (SP-DOCK) and a similarity-guided docking (SG-DOCK). The basic idea is
to maximally exploit the structural information about the ligand binding m
ode present in cases where ligand-bound protein structures are available, i
nformation that is usually ignored in standard docking procedures. SP-DOCK
and SG-DOCK have been derived as modified versions of the program DOCK 4.0,
where the similarity program MIMIC acts as a module for the calculation of
similarity indices that correct docking energy scores at certain steps of
the calculation. SP-DOCK applies similarity corrections to the set of ligan
d orientations at the end of the ligand incremental construction process, p
enalizing the docking energy and, thus, having only an effect on the relati
ve ordering of the final solutions. SG-DOCK applies similarity corrections
throughout the entire ligand incremental construction process, thus affecti
ng not only the relative ordering of solutions but also actively guiding th
e ligand docking. The performance of SP-DOCK and SG-DOCK for binding mode a
ssessment and molecular database screening is discussed. When applied to a
set of 32 thrombin ligands for which crystal structures are available, SG-D
OCK improves the average RMSD by ca, 1 fi when compared with DOCK. When tho
se 32 thrombin ligands are included into a set of 1,000 diverse molecules f
rom the ACD, DIV, and WDI databases, SP-DOCK significantly improves the ret
rieval of thrombin ligands within the first 10% of each of the three databa
ses with respect to DOCK, with minimal additional computational cost. In al
l cases, comparison of SP-DOCK and SG-DOCK results with those obtained by D
OCK and MIMIC is performed. (C) 2000 Wiley-Liss, Inc.