Machine Learning-Driven Data Valuation for Optimizing High-Throughput Screening Pipelines.

Area of Science:

• Drug Discovery and Development
• Computational Chemistry
• Machine Learning in Bioinformatics

Background:

• High-throughput screening (HTS) is critical for identifying bioactive compounds in drug discovery.
• Current HTS methods face challenges with data interpretation, false positives/negatives, and imbalanced datasets.
• Computational efficiency is a key consideration for large-scale screening processes.

Purpose of the Study:

• To introduce and evaluate a novel data valuation approach for enhancing drug discovery pipelines.
• To improve active learning strategies in compound library screening.
• To accurately identify true and false positives and important inactive samples in HTS data.

Main Methods:

• Application of data valuation techniques to assess the importance of data points in HTS.
• Utilizing machine learning models for accurate classification of biological activity versus assay artifacts.
• Implementing importance-based methods for batch screening and undersampling imbalanced datasets.

Main Results:

• Demonstrated effectiveness of importance-based methods for more efficient batch screening, reducing the need for extensive HTS.
• Machine learning models successfully differentiated true biological activity from assay artifacts.
• Importance undersampling improved HTS dataset balancing and machine learning performance without losing critical inactive samples.

Conclusions:

• Data valuation offers a powerful tool to enhance the efficiency and accuracy of drug discovery pipelines.
• The proposed methods streamline the identification of bioactive compounds and improve the reliability of HTS data.
• These advancements have the potential to significantly accelerate the drug development process.