Deconvoluting low yield from weak potency in direct-to-biology workflows with machine learning

Area of Science:

• Medicinal Chemistry
• Drug Discovery and Development
• Computational Chemistry

Background:

• High-throughput biological evaluation of small molecules is crucial for efficient drug discovery.
• Direct-to-biology (D2B) screening accelerates compound evaluation by omitting purification but requires high reaction yields.
• Low yields in D2B assays can lead to misinterpretation of results, mistaking low potency for false negatives.

Purpose of the Study:

• To develop a machine learning model to deconvolve low yields from low potency in D2B screening.
• To identify false negatives in biological assays caused by insufficient compound production.
• To validate the machine learning approach in identifying potent SARS-CoV-2 main protease inhibitors.

Main Methods:

• Implementation of a machine learning-based yield-assay deconfounder.
• Application of the deconfounder to analyze D2B screening data.
• Validation using SARS-CoV-2 main protease inhibitor screening and in silico analysis.

Main Results:

• The machine learning model successfully distinguished between low yield and low potency, identifying true active compounds.
• Promising SARS-CoV-2 main protease inhibitors with nanomolar activity were identified, comparable to standard D2B workflows.
• The framework demonstrated utility in broad in silico screens for discovering compounds with D2B assay-comparable potency.

Conclusions:

• Machine learning can effectively deconvolve yield and potency issues in D2B screening, improving accuracy.
• This approach enhances the identification of viable drug candidates, particularly in PROTAC design.
• The developed framework offers a robust method for efficient and reliable small molecule evaluation in drug discovery.