Bayesian Multi-Plate High-Throughput Screening of Compounds.

Area of Science:

• Computational chemistry
• Statistical modeling in drug discovery
• Bioinformatics

Background:

• High-throughput screening (HTS) is crucial for identifying drug candidate hits from vast compound libraries.
• Current statistical methods, like the B-score, analyze plates individually, neglecting valuable cross-plate correlations.
• This limitation can reduce the sensitivity and specificity of hit identification in drug discovery pipelines.

Purpose of the Study:

• To introduce a novel Bayesian nonparametric statistical framework for analyzing HTS data across multiple plates.
• To enhance the identification of statistically significant compound hits by leveraging shared information among plates.
• To provide a robust method for hit identification and false discovery rate control in large-scale HTS.

Main Methods:

• Development of a Bayesian nonparametric modeling approach for HTS data analysis.
• Simultaneous identification of candidate hits from multiple compound plates, pooling statistical strength.
• Implementation as an efficient R extension package (BHTSpack) for scalability to large datasets.

Main Results:

• The proposed framework demonstrated superior sensitivity and specificity in hit identification compared to B-score and R-score methods.
• Significant improvements were observed even at low hit rates, indicating enhanced detection power.
• The method accommodates arbitrary distributions of compound activities and various plate geometries.

Conclusions:

• The Bayesian nonparametric approach offers a principled and more robust statistical framework for HTS hit identification.
• This method effectively utilizes cross-plate correlations to improve the accuracy of identifying potential drug candidates.
• The BHTSpack R package provides a scalable and efficient tool for modern drug discovery research.