AiKPro: deep learning model for kinome-wide bioactivity profiling using structure-based sequence alignments and...

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Kinase inhibitors are vital for treating diseases, but their development is hindered by kinase structural similarities.
• Efficient kinome-wide bioactivity profiling is crucial for understanding kinase functions and identifying selective inhibitors.

Purpose of the Study:

• To develop AiKPro, a deep learning model for predicting kinase-ligand binding affinities.
• To assess AiKPro's performance in predicting interactions for known and novel kinases and compounds.

Main Methods:

• AiKPro integrates structure-validated multiple sequence alignments and 3D molecular conformer ensemble descriptors.
• An attention-based mechanism is employed to model kinase-ligand interactions.
• Model performance was evaluated using Pearson's correlation coefficients and kinase activity profiling.

Main Results:

• AiKPro achieved high prediction accuracy, with Pearson's correlation coefficients of 0.88 (test set) and 0.87 (untrained compounds).
• The model demonstrated robustness and good kinase-activity profiling across the kinome.
• These results indicate AiKPro's potential for identifying novel interactions and selective inhibitors.

Conclusions:

• AiKPro offers a powerful computational approach for predicting kinase-ligand binding affinities.
• The model facilitates the discovery of novel, selective kinase inhibitors.
• AiKPro can guide rational drug design for kinase-targeted therapies.