DEL-Dock: Molecular Docking-Enabled Modeling of DNA-Encoded Libraries.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• DNA-encoded libraries (DEL) accelerate hit identification by screening vast molecular libraries.
• DEL screens measure binding affinity via DNA sequencing, but data is often noisy.
• Current computational models for DEL data primarily use 2-D representations.

Purpose of the Study:

• To develop a novel computational paradigm, DEL-Dock, for denoising DEL count data.
• To improve the prediction of molecule enrichment scores and binding affinities from DEL screens.
• To leverage 3-D spatial information for more accurate binding modality assessment.

Main Methods:

• DEL-Dock integrates ligand-based descriptors with 3-D spatial information from docked protein-ligand complexes.
• The model learns from the binding modality using 3-D structural data.
• It utilizes probabilistic formulations for denoising DEL count data.

Main Results:

• DEL-Dock effectively denoises DEL count data, outperforming prior methods in correlating predictions with experimental binding affinities.
• The model demonstrates improved prediction of molecule enrichment scores.
• Trained solely on DEL data, DEL-Dock implicitly learns to select accurate docking poses without external structural information.

Conclusions:

• DEL-Dock represents a significant advancement in computational analysis of DEL screening data.
• Incorporating 3-D spatial information enhances the accuracy of binding affinity predictions.
• The model's ability to perform pose selection without crystal structures offers a more efficient drug discovery pipeline.