Multi-objective active machine learning rapidly improves structure-activity models and reveals new protein-protein...

Area of Science:

• Computational Chemistry
• Machine Learning in Drug Discovery
• Medicinal Chemistry

Background:

• Protein-protein interactions (PPIs) are critical in biological processes.
• Inhibiting the CXCR4-CXCL12 interaction is a promising anti-cancer strategy.
• Traditional drug discovery methods can be time-consuming and resource-intensive.

Purpose of the Study:

• To apply a multi-objective active learning scheme for identifying small molecule inhibitors.
• To target the protein-protein interaction between CXC chemokine receptor 4 (CXCR4) and CXCL-12 (SDF-1).
• To enhance the efficiency of drug discovery through adaptive experimental design.

Main Methods:

• Utilized active machine learning for experimental design and compound selection.
• Developed an adaptive structure-activity model.
• Employed a multi-objective compound selection function for balanced optimization.
• Screened large compound repositories and virtual libraries.

Main Results:

• Successfully identified small molecules with desired inhibitory activity against the CXCR4-CXCL12 interaction.
• Demonstrated that active learning continuously improved the structure-activity model.
• Showcased the ability to rapidly deliver new molecular structures with inhibitory potential.
• Highlighted the focus on informative compounds for efficient model adjustment.

Conclusions:

• Active learning is a validated approach for prospective ligand discovery.
• Adaptive, focused screening accelerates the identification of potent inhibitors.
• This strategy optimizes the use of large compound libraries for targeted drug development.