3D-Sensitive Encoding of Pharmacophore Features.

Area of Science:

• Computational chemistry
• Cheminformatics
• Structural biology

Background:

• Comparing three-dimensional (3D) molecules and binding sites is crucial in drug discovery and structural biology.
• Existing methods often rely on computationally intensive superposition or complex feature representations.

Purpose of the Study:

• To introduce an efficient, overlay-free method for ranking 3D molecules and comparing binding sites based on pharmacophore features.
• To develop a computationally scalable approach for large-scale virtual screening and structural analysis.

Main Methods:

• A novel encoding scheme for pharmacophore features with only two fittable parameters, designed to be sparse and low-dimensional.
• Extension of the method to compare protein-ligand binding sites by incorporating an additional parameter.
• Benchmarking on LIT-PCBA for ligand-based virtual screening and sc-PDB/Vertex datasets for binding site comparison.

Main Results:

• The proposed method outperforms existing overlay-based and fingerprint methods in ligand-based similarity searches.
• It demonstrates competitive performance against state-of-the-art methods for binding site comparisons.
• Achieves high throughput, ranking up to 374,000 ligands/s or 132,000 binding sites/s on a single CPU core.

Conclusions:

• The developed overlay-free pharmacophore feature encoding provides an efficient and scalable solution for 3D molecule and binding site comparison.
• This method significantly advances virtual screening and structural analysis capabilities in cheminformatics.
• The open-source software 'AutoCorrelation of Pharmacophore Features' (ACP4) is released to facilitate further research and application.