Generalized modeling of enzyme-ligand interactions using proteochemometrics and local protein substructures

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Protein-ligand interactions are crucial for drug discovery.
• Current proteochemometrics methods often require closely related protein families.
• Accurate modeling necessitates considering both protein structure and ligand properties.

Purpose of the Study:

• To develop a generalized proteochemometrics approach for modeling protein-ligand binding affinity.
• To represent diverse proteins using a library of local substructures.
• To investigate the relationship between local protein structures and ligand binding.

Main Methods:

• Utilized linear regression and rule-based learning.
• Combined local protein substructures with ligand chemical descriptors.
• Employed cross-validation and independent test sets for evaluation.

Main Results:

• The new approach outperformed models using only protein or ligand descriptors.
• Models demonstrated generalization across proteins with varying sequence and structure.
• Identified specific local substructures associated with strong and weak binding.

Conclusions:

• Local substructure analysis enables modeling of protein-ligand interactions in diverse proteins.
• This alignment-independent method expands proteochemometrics to entire proteomes.
• The findings offer insights into the structural basis of enzyme-ligand binding specificity.