Binding interaction of acetylcholinesterase with steroidal pregnanes: insight from machine learning and atomistic simulation

ABSTRACT
Acetylcholinesterase (AChE) inhibition is a key strategy in the treatment of Alzheimer’s disease and other
neurodegenerative disorders. While pregnane-based compounds have been suggested as AChE inhibitors,
their mechanism of action remains unclear. This study employed machine learning (ML) and molecular
modeling to probe the molecular interaction of AChE with steroidal pregnanes. The ML models were
trained and validated on AChE bioactivity datasets to predict pIC50 and pKi values of small-molecule
compounds. Among the models tested, the Random Forest Regressor demonstrated superior performance
and was used to identify pregnanes with pIC50�5 and pKi � 7 as promising inhibitors. Molecular
docking revealed strong molecular interactions between AChE and several pregnanes, particularly 21-[(3-
Hydroxy-2-naphthyl)oxy]pregnane-2-one. This compound interacted with critical sub-sites within the
AChE binding gorge, including the catalytic active site, peripheral anionic site, oxyanion hole, and anionic
sub-site, through multiple hydrogen bonds and hydrophobic interactions. Molecular dynamics simulations
over 100 ns indicated structural stability and conformational flexibility of the representative AChE-pregnane
complexes as indicated by the dynamic parameters and cluster patterns. The Molecular Mechanics
with Generalized Born Surface Area free energy analysis confirmed strong binding affinities, while
residual energy decomposition provided insights into key residue contributions. Additionally, the pregnanes
demonstrated favorable blood-brain barrier permeability and other drug-like properties, suggesting
their potential as neurotherapeutic agents. Given their predicted bioactivity, strong interactions with
AChE, and drug-like properties, the identified pregnanes warrant further optimization and experimental
evaluation for the development of safe and effective AChE inhibitors.
List of abbreviations: AChE: Acetylcholinesterase; ADMET: Absorption, Distribution, Metabolism,
Excretion, and Toxicity; BBB: Blood-Brain Barrier; CAS: Catalytic Active Site; CNS Central Nervous
System; HBA: Hydrogen Bond Acceptor; HBD: Hydrogen Bond Donor; IC50: Half-Maximal Inhibitory
Concentration; Ki: Inhibition Constant; LogP: Partition Coefficient (Lipophilicity Indicator); MD:
Molecular Dynamics; ML: Machine Learning; MM: Molecular Mechanics; MMGBSA: Molecular
Mechanics/Generalized Born Surface Area; MW: Molecular Weight; NPT: Constant Number, Pressure,
and Temperature (MD Simulation Condition); NVT: Constant Number, Volume, and Temperature (MD
Simulation Condition); PAS: Peripheral Anionic Site; PDB: Protein Data Bank; pIC50: Negative Logarithm
of IC50; pKi: Negative Logarithm of Ki; QSAR: Quantitative Structure-Activity Relationship; RCSB:
Research Collaboratory for Structural Bioinformatics; RFR: Random Forest Regressor; RMSD: Root Mean
Square Deviation; RMSF: Root Mean Square Fluctuation; RoG: Radius of Gyration; RO5: Rule of Five
(Lipinski’s Drug-Likeness Criteria); SASA: Solvent Accessibility Surface Area; SBVS: Structure-Based Drug
Design; VDW: Van der Waals Interactions; XGBoost: Extreme Gradient Boosting Algorithm; DG: Free
Energy of Binding