In Silico Fragment-Based Drug Discovery: Setup and Validation of a Fragment-to-Lead Computational Protocol Using S4MPLE

This paper describes the use and validation of S4MPLE in Fragment-Based
Drug Design (FBDD)a strategy to build drug-like ligands starting
from small compounds called fragments. S4MPLE is a conformational
sampling tool based on a hybrid genetic algorithm that is able to
simulate one (conformer enumeration) or more molecules (docking).
The goal of the current paper is to show that due to the judicious
design of genetic operators, S4MPLE may be used without any specific
adaptation as an in silico FBDD tool. Such fragment-to-lead evolution
involves either growing of one or linking of several fragment-like
binder(s). The native ability to specifically “dock”
a substructure that is covalently anchored to its target (here, some
prepositioned fragment formally part of the binding site) enables
it to act like dedicated de novo builders and differentiates it from
most classical docking tools, which may only cope with non-covalent
interactions. Besides, S4MPLE may address growing/linking scenarios
involving protein site flexibility, and it might also suggest “growth”
moves by bridging the ligand to the site via water-mediated interactions
if H2O molecules are simply appended to the input files.
Therefore, the only development overhead required to build a virtual
fragment→ligand growing/linking strategy based on S4MPLE were
two chemoinformatics programs meant to provide a minimalistic management
of the linker library. The first creates a duplicate-free library
by fragmenting a compound database, whereas the second builds new
compounds, attaching chemically compatible linkers to the starting
fragments. S4MPLE is subsequently used to probe the optimal placement
of the linkers within the binding site, with initial restraints on
atoms from initial fragments, followed by an optimization of all kept
poses after restraint removal. Ranking is mainly based on two criteria:
force-field potential energy and RMSD shifts of the original fragment
moieties. This strategy was applied to several examples from the FBDD
literature with good results over several monitored criteria: ability
to generate the optimized ligand (or close analogs), good ranking
of analogs among decoy compounds, and accurate predictions of expected
binding modes of reference ligands. Simulations included “classical”
covalent growing/linking, more challenging ones involving binding
site conformational changes, and growth with optional recognition
of putatively favorable water-mediated interactions.