Cell Painting-based bioactivity prediction boosts high-throughput screening hit-rates and compound diversity

Area of Science:

• Computational biology
• Drug discovery
• Machine learning

Background:

• Early drug discovery is resource-intensive, requiring identification of active compounds.
• Morphological profiling offers a potential method to streamline compound screening.
• Cell Painting data provides rich cellular information for phenotypic analysis.

Purpose of the Study:

• To investigate the use of deep learning on Cell Painting data for predicting compound bioactivity.
• To assess the performance of models across a diverse set of 140 assays.
• To determine if brightfield images alone can achieve high prediction accuracy.

Main Methods:

• Utilized deep learning models trained on Cell Painting images and single-concentration activity readouts.
• Predicted compound activity across 140 different assays.
• Analyzed model performance using ROC-AUC scores and validated predictions experimentally.

Main Results:

• Achieved an average ROC-AUC of 0.744 ± 0.108, with 62% of assays reaching ≥0.7.
• Demonstrated high prediction performance, sometimes achievable with only brightfield images.
• Confirmed robustness across various assay types, technologies, and target classes, especially cell-based assays and kinase targets.

Conclusions:

• Cell Painting-based bioactivity prediction is a reliable method for diverse targets and assays.
• This approach can significantly reduce screening campaign sizes, saving time and resources.
• Models trained on Cell Painting data enhance hit rates and scaffold diversity in drug discovery.