Kinome-Wide Profiling Prediction of Small Molecules

Area of Science:

• Computational chemistry and drug discovery
• Bioinformatics and cheminformatics
• Pharmacology and kinase inhibitor research

Background:

• Large-scale kinase profiling data is now available, enabling the development of predictive activity models.
• Proteochemometric (PCM) modeling leverages compound and protein descriptors for bioactivity prediction.
• Existing quantitative structure-activity relationship (QSAR) models have limitations in predicting activity for novel kinases and compounds.

Purpose of the Study:

• To explore the potential of PCM for large-scale kinase activity prediction across the entire human kinome.
• To assess PCM's applicability to novel compounds and clinically relevant kinase mutants.
• To compare PCM's predictive performance against traditional kinase QSAR models.

Main Methods:

• Development and application of proteochemometric (PCM) models using extensive kinase profiling data.
• Rigorous validation of PCM models on left-out kinases and novel compounds.
• External validation of PCM models on clinically relevant mutant kinases, including those in the ATP binding site.

Main Results:

• PCM models demonstrate high predictive power for kinases not included in the training set.
• PCM significantly outperforms individual kinase QSAR models when predicting activity for new compounds.
• External validation shows excellent predictive accuracy for mutant kinases, even with mutations across the ATP binding site.

Conclusions:

• Proteochemometric modeling is a powerful and versatile approach for predicting kinase activity at a large scale.
• PCM effectively extrapolates bioactivity to unexplored kinases and novel chemical entities.
• The predictive capability extends to clinically relevant kinase mutants, offering potential for drug resistance studies.