LEADD

LEADD performs goal-directed de novo molecular design to optimize molecular properties while enforcing synthetic feasibility.

Key Features:

• Fragment-based molecular graphs: Represents molecules as graphs composed of molecular fragments.
• Pairwise atom type compatibility rules: Uses knowledge-based pairwise atom type compatibility rules that constrain which bonds can form between fragments.
• Reference library extraction: Extracts fragments from a reference library of drug-like molecules to derive compatibility rules.
• Constraint-enforcing genetic operators: Implements novel genetic operators tailored to enforce compatibility constraints during evolution.
• Lamarckian evolutionary mechanism: Adapts reproductive behavior of molecules dynamically based on fitness within the objective function.
• Goal-directed optimization: Performs optimization of molecular properties under user-specified objective functions.
• Benchmarked performance: Demonstrated improved fitness and predicted synthetic accessibility relative to virtual screening and other evolutionary algorithms on standardized benchmark suites.

Scientific Applications:

• De novo drug discovery: Generates novel candidate molecules for drug discovery while prioritizing synthetic feasibility.
• Property optimization: Optimizes key molecular properties (fitness metrics) in a goal-directed manner.
• Synthetic accessibility-driven design: Produces molecules constrained by compatibility rules to enhance practical synthetic tractability.
• Method comparison and benchmarking: Serves as a comparative approach against virtual screening and other evolutionary algorithms using standardized benchmark suites.

Methodology:

Represents molecules as fragment-based graphs; derives knowledge-based pairwise atom type compatibility rules from a reference library of drug-like molecules; applies novel genetic operators that enforce compatibility constraints within an evolutionary algorithm incorporating a Lamarckian adaptation of reproductive behavior; evaluated on standardized benchmark suites against virtual screening and other evolutionary algorithms.