Convex-PL: a novel knowledge-based potential for protein-ligand interactions deduced from structural databases using...

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Accurate protein-ligand scoring functions are crucial for drug discovery.
• Existing methods often rely on predefined functional forms, limiting their flexibility.
• Training data generation can introduce biases, affecting generalizability.

Purpose of the Study:

• To introduce Convex-PL, a novel, free-shape, distance-dependent protein-ligand scoring function.
• To develop an optimization approach that does not impose a functional form on the scoring function.
• To create a generally applicable scoring function for diverse structural ensembles.

Main Methods:

• Decomposing the scoring function into a polynomial basis.
• Using a convex formulation to deduce expansion coefficients from a structural knowledge base.
• Employing constant RMSD rigid-body deformations for training data instead of docking-generated poses.

Main Results:

• Convex-PL demonstrated superior performance compared to 20 other methods in the CASF 2013 benchmark.
• The scoring function achieved high accuracy on D3R Grand Challenge 2 submissions.
• The method shows general applicability across different docking approaches.

Conclusions:

• Convex-PL offers a flexible and accurate approach to protein-ligand scoring.
• The novel training strategy enhances the generalizability of the scoring function.
• This method advances computational drug discovery by improving scoring function performance.