ABCpred

ABCpred predicts acetylcholinesterase (AChE) and butyrylcholinesterase (BChE) inhibitory activity using machine-learning classification models to support discovery of cholinesterase inhibitors for Alzheimer's disease research.

Key Features:

• Structure-Activity Relationship Model: A large-scale classification SAR model identifies key substructures that govern cholinesterase inhibitory activity.
• Data Source and Molecular Descriptors: Non-redundant datasets of 985 AChE and 1056 BChE compounds from the ChEMBL database are characterized using 12 sets of molecular fingerprints.
• Predictive Modeling with Random Forest: Predictive classification models are developed using the random forest algorithm.
• Model Evaluation and Interpretability: Performance is assessed using the Matthews correlation coefficient (MCC) with five-fold cross-validation and independent test sets; the SubstructureCount fingerprint achieved cross-validated MCCs of 0.76 (AChE) and 0.82 (BChE) and test MCCs of 0.73 and 0.97, respectively.
• Feature Interpretation: Chemical features identified as critical for inhibition include aromatic ring systems, heterocyclic nitrogen-containing compounds, and amines.

Scientific Applications:

• Lead identification: Predicts AChE and BChE inhibitory activity to prioritize promising lead compounds for cholinesterase inhibitor development.
• Drug discovery support: Provides high-accuracy predictions to support compound prioritization in Alzheimer's disease drug discovery efforts.
• Mechanistic insight and experimental guidance: Interpretable substructure and fingerprint insights inform understanding of molecular determinants of inhibition and guide experimental validation and optimization.

Methodology:

Models were built from non-redundant ChEMBL datasets (985 AChE, 1056 BChE), encoded with 12 sets of molecular fingerprints including SubstructureCount, trained as classification SAR models using the random forest algorithm, and evaluated by five-fold cross-validation and independent test sets using the Matthews correlation coefficient (MCC).