Page 107 - IMO-2-2

Innovative Medicines & Omics Modeling Aurora-B kinase inhibitors

2.2.1. Ligand preparation atom-based QSAR model, in which the ligand structural

The 2D structures of compounds were imported into the components were represented by van der Waals atomic
Develop Common Pharmacophore Hypotheses (CPHs) models. Atoms occupying the same region were categorized
panel. The structures were minimized and geometrically into six classes:
refined using the LIGPREP module, which neutralized • Hydrogen-bond donors (D) – Atoms such as nitrogen,
the ionized structures to a pH 7 (neutral) and generated oxygen, phosphorus, and sulfur bonded to a hydrogen
possible stereoisomers. 32,33 Conformers were generated atom.
using the rapid torsional angle search method (ConfGen) • Hydrophobic or nonpolar groups (H) – Carbon,
with distance-dependent dielectric solvation treatment and hydrogen attached to carbon, and halogens.
OPLS–2005 force field incorporated in PHASE. Molecular • Negatively charged groups (N) – Atoms or functional
docking simulations were performed using an implicit groups carrying a formal negative charge.
solvent model with a distance-dependent dielectric • Positively charged groups (P) – Atoms or functional
(GB/SA) approach. The simulations employed a cutoff of 1 groups carrying a formal positive charge.
Å RMSD for interactions and consisted of 1000 iterations, • Electron-withdrawing groups (W) – Atoms such
with water as the implicit solvent. For each structure, a as nitrogen and oxygen, including those that act as
maximum of 1000 conformers were generated using 100 hydrogen-bond acceptors.
steps of pre-process minimizations and 50 steps of post- • Miscellaneous groups (X) – All other atoms and
process minimizations. The maximum energy difference functional groups that do not fit into the above
for a set of conformers of each molecule is 10 kcal/mol. The categories.
active and inactive ligands were assigned based on activity The partial least square (PLS) regression was carried
threshold values. out for QSAR modeling in PHASE, with a maximum of
N/5 factors (where, N = number of ligands in the training
2.2.2. Creating pharmacophore sites
set). The model’s accuracy improved with an increasing
Pharmacophore features were defined to create sites for number of PLS factors until overfitting was observed. 34,35
all ligands. PHASE provided six pharmacophore features: Three PLS factors were generated for all hypotheses with a
hydrogen bond acceptor (A), hydrogen bond donor (D), grid spacing of 1 Å and the best model was selected based
hydrophobic group (H), negatively ionizable (N), positively on statistical parameters, such as R , Q , SD, RMSE, F,
2
2
ionizable (P), and aromatic ring (R). All six features were Pearson R, and stability values for virtual screening.
utilized in pharmacophore site creation.
2.3. Virtual screening
2.2.3. Finding common pharmacophore
The validated hypothesis was used as a query to search for
Common pharmacophores were identified from the set novel Aurora-B inhibitors. The National Cancer Institute
of variants using a tree-based partition technique with a (NCI) and Maybridge databases (https://dtp.cancer.gov/
maximum depth of 5 and a minimum intersite distance of databases_tools/bulk_data.htm, https://www.thermofisher.
2.0 Å. The initial and final box sizes were set to 32.0 Å and in/chemicals/en/forms/maybridge-downloads.html)
1.0 Å, respectively, ensuring all active compounds were were explored to identify potential chemical structures.
matched. CPHs were generated by varying the maximum Hit molecules were further filtered based on in silico
and minimum number of sites and the number of matching pharmacokinetic properties – absorption, distribution,
active groups. metabolism, and excretion (ADME) – using the QIKPROP
module, ensuring compliance with Lipinski’s rule of five
2.2.4. Scoring hypotheses for drug-likeness. 36-38 The molecules with drug-likeness
The generated CPHs were examined using the scoring were subjected to molecular docking to find the best-
procedure to find the best alignment of active molecules. fit interactions within the active site of the Aurora-B
The scoring process ranked hypotheses based on distinct protein. The Glide software offers three distinct levels of
features. The hypotheses table was used to choose the most docking methodologies: high throughput virtual screening
appropriate hypothesis for further investigation. (HTVS), standard precision (SP), and extra precision (XP).
Initially, HTVS docking was employed to predict protein-
2.2.5. Building QSAR model ligand binding modes and rank ligands utilizing empirical
For QSAR modeling, the dataset was divided into a scoring functions. The top-ranked ligands from HTVS
training set (70%) and a test set (30%) based on the underwent SP docking for further refinement. Finally,
selected hypothesis. PHASE provides an atom-based and the most promising molecules were subjected to XP
pharmacophore feature-based QSAR model. We used an docking, which employs an anchor-and-grow algorithm

Volume 2 Issue 2 (2025) 101 doi: 10.36922/imo.6547