Comparison of Chemical Structure and Cell Morphology Information for Multitask Bioactivity Predictions.

Area of Science:

• Computational chemistry
• Chemical biology
• Drug discovery

Background:

• Understanding compound mechanism-of-action (MoA) and predicting drug targets are crucial for small-molecule drug discovery.
• Cellular imaging and chemical structure data offer distinct information for bioactivity prediction.

Purpose of the Study:

• To compare the predictive performance of cell morphology data versus chemical structure information for bioactivity prediction.
• To evaluate the utility of these data types for predicting compound effects on various biological targets.

Main Methods:

• Utilized bioactivity data from the ExCAPE database and cell imaging features from the Cell Painting dataset.
• Employed the multitask Bayesian matrix factorization (BMF) approach, Macau, to analyze extended connectivity fingerprints (ECFPs) and image-derived features.
• Compared BMF Macau performance with random forest (RF) for target prediction.

Main Results:

• BMF Macau and RF showed similar performance using ECFPs.
• BMF Macau significantly outperformed RF when using cell image data (71% of targets).
• High predictive performance (AUC > 0.8) was achieved for 45% of targets with ECFPs and 40% with image data, with some targets showing complementary prediction strengths.

Conclusions:

• Both cell morphology and chemical structure information contribute valuable, partially complementary data for predicting compound bioactivity.
• These findings support the integration of diverse data types for enhanced in silico MoA analysis and drug target identification.