Ligand-Based Computation of HIV-1 Integrase Inhibition Strength within a Series of β-ketoamide derivatives
A continuous demand exists for novel bioactive molecules. When
a lead structure has been discovered and looks promising for further development,
series of analogues will be made. Normally, the synthesis of many
compounds is required to improve on the activity, or to keep good activity
while optimising other properties of relevance. A computational model that
accurately predicts the activity of derivatives before their
synthesis is beneficial to the speed and cost of lead optimisation. It
can be advantageous when such a model does not require information on the
target protein structure.
A conformational analysis was performed on 201 ketoamide ester
derivatives that inhibit HIV integrase. The derivatives were aligned to the lowest
energy conformer of the most potent inhibitor with the SEAL method using our
NSGASEAL application. Five
CoMSIA fields were computed for each compound taking into account steric,
polarisability, charge, H-bond acceptor, and H-bond donor properties. A
model for integrase-inhibitor interaction was derived by PLS regression.
An sample application
of a web-based front-end for the NSGASEAL driven model can be tried on line.
Reprints of this paper can be requested with our information request form or from the journal IEJMD directly.
The predictivity of the model was tested by scrambling the data,
leave-n-out experiments and applying the model to a ketoamide acid series
of integrase inhibitors. In order to elucidate the binding mode of the
inhibitors, the model fields were subsequently mapped on a crystal structure
of the integrase enzyme.
The CoMSIA model derived from the 201 ketoamide ester derivatives has an
r2 of 0.75. The resulting fields of the molecular properties required
for strong inhibition can be qualitatively understood. Scrambling the
data prohibited the derivation of a predictive model. The models derived from 100
derivatives when applied to the remaining 101 compounds, resulted in
a prediction with an absolute deviation of 0.28 log10 unit/compound.
The accuracy of prediction when the model was applied to 74 ketoamide acids was
0.42 log10 unit/compound. Mapping the model onto the integrase
enzyme did not lead to an obvious binding mode.
The predictivity of our model clearly allows for guiding
the synthesis of novel analogues. The approach holds its predictive value when applied to a
different series, though to a lesser extent. The geometry of integrase-inhibitor
binding is not very well understood at the present time, which emphasizes
the advantage of an approach that does not require this knowledge for
the design of novel active compounds.
