Identification of Compounds That Interfere with High-Throughput Screening Assay Technologies.

Area of Science:

• Drug discovery and development
• Computational chemistry
• Assay development

Background:

• High-throughput screening (HTS) campaigns face challenges identifying assay technology interference compounds (CIATs).
• CIATs produce false readouts, leading to wasted resources on non-viable hits.
• Existing CIAT identification methods often rely on statistical analysis of historical data, neglecting experimental validation.

Purpose of the Study:

• To develop a machine-learning (ML) model for predicting CIATs across three distinct assay technologies.
• To improve the accuracy and scope of CIAT identification beyond current methods.
• To leverage well-curated datasets for robust predictive modeling.

Main Methods:

• Training an ML model using known CIATs and non-CIATs (NCIATs).
• Utilizing 2D structural descriptors to characterize compounds.
• Comparing the ML model's predictions against structure-independent models (BSF) and substructural filters (PAINS).

Main Results:

• Successful prediction of CIATs for both known and novel chemical entities.
• The ML model provided a broader set of predicted CIATs compared to BSF and PAINS filters.
• Demonstrated the power of well-curated datasets in building effective predictive models, even with limited data size.

Conclusions:

• The developed ML model offers a powerful tool for predicting CIATs in HTS.
• This approach enhances the reliability of HTS campaigns by proactively identifying interfering compounds.
• The study highlights the value of experimental data and ML in addressing critical challenges in drug discovery.