Compound Functional Prediction Using Multiple Unrelated Morphological Profiling Assays.

Area of Science:

• Drug discovery and development
• Cellular biology
• Computational biology

Background:

• Phenotypic cell-based assays are crucial for identifying novel therapeutic drugs by screening diverse targets.
• Identifying the mechanism of action (MOA) for hit compounds remains a significant bottleneck in automated drug discovery pipelines.
• Current methods like cell painting assays use multiplexed fluorescent dyes, but their MOA prediction accuracy is limited.

Purpose of the Study:

• To investigate whether combining simple, unrelated cell-based assays can improve mechanism of action (MOA) prediction accuracy.
• To determine if a single assay's multiplexed measurements are highly correlated and if nuclei staining alone is sufficient for MOA prediction.
• To develop a more biologically and technically relevant approach for predicting drug targets from phenotypic screens.

Main Methods:

• Assessed the impact of adding fluorescent dyes to single assays on MOA prediction accuracy.
• Evaluated the predictive power of monitoring only the nuclei stain in cell-based assays.
• Developed and trained an ensemble classifier using combined data from multiple, simple cell-based assays.

Main Results:

• Monitoring only the nuclei stain in a single assay achieved compelling MOA prediction accuracy.
• Multiplexed measurements from different dyes within a single assay were found to be highly correlated.
• Combining data from simple, unrelated cell-based assays significantly enhanced MOA prediction capabilities.

Conclusions:

• Nuclei staining alone can reflect the general cellular state and is a strong predictor of MOA.
• Combining simple, diverse cell-based assays offers a more robust and relevant strategy for MOA prediction than complex single assays.
• This approach enables the effective reuse of historical screening data to train ensemble classifiers for prioritizing drug candidates.