SCAM detective

SCAM detective predicts small colloidally aggregating molecules (SCAMs) in chemical libraries used in high-throughput screening (HTS) by estimating detergent-sensitive aggregation propensity to reduce false positives.

Key Features:

• QSIR models: Quantitative structure-interference relationship (QSIR) models predict detergent-sensitive aggregation from chemical structure.
• Machine learning implementation: Machine learning classifiers underlie the QSIR models for SCAM prediction.
• Assay-specific focus: Models are specifically developed for the AmpC β-lactamase assay and a cruzain inhibition assay.
• Validation across HTS data: Models were developed and validated using data from multiple HTS campaigns under varied assay conditions and screening concentrations.
• Performance improvement: Prediction accuracy is reported to improve by approximately 53% for the β-lactamase assay and approximately 46% for the cruzain assay versus previously published methods.
• Consideration of assay variables: Model development accounts for assay conditions and screening concentrations.

Scientific Applications:

• False-positive reduction in HTS: Identification of detergent-sensitive aggregators to decrease assay false positives in HTS campaigns.
• Aggregation counter-screening: Supports use of AmpC β-lactamase as a counter-screen to detect aggregation-related interference.
• Model development for interference mechanisms: Provides data and approach to inform QSIR model development for other assay interference mechanisms beyond aggregation.
• Improving HTS reliability: Enhances reliability and interpretation of screening results across diverse assay conditions and concentrations.

Methodology:

QSIR models implemented as machine learning classifiers were trained and validated on HTS datasets, including AmpC β-lactamase and cruzain assay data, across varied assay conditions and screening concentrations to predict detergent-sensitive aggregation.