Evolutionary chemical binding similarity approach integrated with 3D-QSAR method for effective virtual screening

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Traditional virtual screening methods are often time-consuming and have low success rates.
• A novel machine learning-based chemical binding similarity model was developed, considering common structural features of molecules binding to related targets.
• This model measures compound resemblance based on binding site similarity to better reflect functional similarities.

Purpose of the Study:

• To evaluate the efficacy of the chemical binding similarity model in virtual screening.
• To compare its performance against conventional structure-based methods.
• To explore its potential for identifying novel drug candidates.

Main Methods:

• Evaluated chemical binding similarity, receptor-based pharmacophore, chemical structure similarity, and molecular docking.
• Tested the chemical binding similarity method on 51 kinase test sets.
• Performed virtual screening on blind datasets for MEK1, EPHB4, and WEE1 using chemical binding similarity and pharmacophore methods.

Main Results:

• The chemical binding similarity method outperformed traditional structural and structure-based methods in identifying active compounds.
• Virtual screening against blind datasets identified novel active compounds: 6/13 for MEK1 and 2/12 for EPHB4.
• The identified molecules often possessed low structural similarity to known inhibitors, indicating novel scaffold discovery.

Conclusions:

• Combining the chemical binding similarity model with 3D-QSAR pharmacophore and molecular docking further enhances virtual screening results.
• The study successfully identified new inhibitors with novel scaffolds, advancing drug discovery.
• The chemical binding similarity model offers a promising approach for more efficient and effective virtual screening.