Computational Prediction of Drug Phenotypic Effects Based on Substructure-Phenotype Associations

Area of Science:

• Computational chemistry and cheminformatics
• Pharmacology and drug discovery
• Machine learning in bioinformatics

Background:

• Accurate prediction of drug phenotypic effects (therapeutic effects and adverse drug reactions/ADRs) is crucial for evaluating new drug candidates (NDCs).
• Current computational methods often rely on overall drug structure or target information, leading to inconsistencies and limited prediction scope.
• A need exists for methods that link local structural features to phenotypic outcomes for more effective NDC evaluation.

Purpose of the Study:

• To develop a computational pipeline for predicting drug phenotypic effects based on substructure-phenotype relationships.
• To enable high-throughput and effective evaluation of the druggability of NDCs using only their structural information.
• To identify risk substructures associated with severe ADRs and analyze drug mechanisms of action (MOA).

Main Methods:

• Constructed quantitative associations between drug substructures, protein domains, and phenotypes (ADRs and Anatomical Therapeutic Chemical Classification System/ATC codes) using L1LOG and L1SVM machine learning models.
• Established substructure-phenotype relationships to quantify drug-phenotype associations.
• Validated predictions through five-fold cross-validation, public databases, literature review, and molecular docking.

Main Results:

• Predicted 83,205 drug-ATC relationships and 306,421 drug-ADR relationships.
• Successfully predicted known effects for Maraviroc (79 ATCs, 269 ADRs) and identified novel substructure-phenotype links, such as SUB215 associated with shock.
• Demonstrated feasibility through case studies and analysis of drug MOA based on established drug-substructure-domain-protein associations.

Conclusions:

• The developed approach enables quantitative prediction of drug phenotypes directly from structural information without prior knowledge.
• This method facilitates the analysis of drug phenotypic mechanisms and accelerates rational drug design.
• The established substructure-phenotype relationships offer a valuable resource for drug discovery and safety assessment.