Deep Learning Approach for the Discovery of Tumor-Targeting Small Organic Ligands from DNA-Encoded Chemical Libraries

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Oncology

Background:

• DNA-Encoded Chemical Libraries (DELs) offer large-scale protein-ligand interaction data for efficient drug discovery.
• Carbonic anhydrase IX (CAIX) is a validated biomarker for tumor hypoxia and clear cell renal cell carcinoma.

Purpose of the Study:

• To develop a novel strategy combining DEL screening and deep learning for identifying tumor-targeting ligands against CAIX.
• To generate novel chemical motifs beyond existing DEL libraries for CAIX inhibition.

Main Methods:

• Instance-level deep learning models were trained on CAIX-screening DEL datasets.
• Machine learning models predicted novel CAIX-binding ligands dissimilar to original DEL elements.
• Hit-to-lead optimization involved prioritizing synthesized analogues based on predicted affinity and feasibility.

Main Results:

• 70% of 152 novel predicted hits showed submicromolar enzymatic inhibition (IC50 < 1 μM).
• Three lead candidates demonstrated accumulation on CAIX-expressing tumor cells.
• The top lead compound exhibited a 5.7 nM K_D and selective in vivo tumor targeting in mice.

Conclusions:

• The synergy between DEL and machine learning accelerates the discovery of novel hits and the development of in vivo tumor-targeting agents.
• This integrated approach successfully translates DEL-derived leads into potential therapeutic candidates for CAIX-positive cancers.
• The study highlights a powerful platform for identifying and optimizing targeted cancer therapies.