Machine Learning-Assisted Iterative Screening for Efficient Detection of Drug Discovery Starting Points.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• High-throughput screening (HTS) is crucial for identifying small molecule leads.
• Increasing assay complexity and large compound libraries pose challenges to traditional HTS.
• Prospective validation of active-learning-guided screening in large-scale projects is limited.

Purpose of the Study:

• To prospectively evaluate machine learning (ML)-assisted iterative HTS in a large-scale drug discovery setting.
• To assess the efficiency and hit recovery of ML-guided screening compared to full HTS.
• To demonstrate the potential of ML to reduce experimental costs while maintaining hit quality.

Main Methods:

• A mass spectrometry-based assay targeting salt-inducible kinase 2 was employed.
• A two million-compound library was screened iteratively using ML guidance over three batches.
• Performance was benchmarked against a parallel full HTS and similarity-based methods.

Main Results:

• Screening only 5.9% of the library yielded 43.3% of all primary active compounds found in full HTS.
• ML-guided screening successfully identified nearly all compound series prioritized by medicinal chemists.
• ML approaches demonstrated superior hit recovery and chemical space coverage compared to similarity-based methods.

Conclusions:

• ML-driven iterative HTS offers a significant reduction in experimental costs for large-scale drug discovery.
• This approach maintains high-quality hit discovery, comparable to or exceeding traditional methods.
• ML-assisted iterative HTS represents a promising strategy to enhance efficiency in modern drug discovery pipelines.