Rise of AI Technologies in Virtual Screening

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery

Background:

• Emergence of AI foundational models for predicting protein-ligand interactions and binding affinities.
• Need for robust methods to analyze large virtual screening datasets.

Purpose of the Study:

• To evaluate the performance of the Boltz-2 AI model on a challenging dataset for protein-ligand interaction prediction.
• To compare Boltz-2 against other rescoring strategies in virtual screening.

Main Methods:

• Utilized a dataset comprising ten ultralarge virtual screening hit lists.
• Employed in vitro binding assays for target validation.
• Challenged the Boltz-2 model on this difficult dataset.

Main Results:

• Boltz-2 emerged as the top-performing classifier.
• Achieved a success rate twice that of any other rescoring strategy.
• Demonstrated straightforward, accurate, efficient, and robust ligand classifications.

Conclusions:

• Boltz-2 offers a significant advancement in predicting protein-ligand interactions.
• Enables accurate, large-scale compound rankings on standard computational resources.
• Paves the way for more efficient drug discovery pipelines.