Improving phosphoproteomic kinase activity inference by contextual refinement of Kinase Library relationships

The accuracy of inferring kinase activity from phosphoproteomic data depends on kinase-specific substrate sets, known as kinase-substrate relationship (KSR) priors. The Kinase Library (KL) broadens the range of potential substrates by incorporating experimentally derived specificity profiles, but its predictions are based on sequence compatibility rather than established intracellular regulation.

To evaluate the effectiveness of KL-expanded KSR priors, researchers analyzed 222 phosphoproteomic samples, each representing perturbations across 78 kinases and encompassing 396 annotated activation or inhibition events. Expanding the KL coverage from 108 to 391 kinases increased the number of scoreable kinases but diminished discrimination power, directional accuracy, and ranking performance.

This reduction in performance was more pronounced in kinases with limited curated substrate data, as the broader expansion also led to increased redundancy among inferred activity profiles.

To enhance the reliability of motif-derived relationships, the researchers introduced constraints to the KL-derived KSRs using predicted protein co-localization and protein-association evidence. Filtering based on protein-association consistently improved the inference performance, reduced activity redundancy, and achieved a more favorable balance between accuracy and coverage compared to co-localization filtering.

By applying the protein-association-filtered prior, the researchers retained activity estimates for 308 kinases with an area under the receiver operating characteristic curve (AUROC) above 0.75. This approach also extended the inference to 200 kinases that were not previously scoreable using curated KSRs. These newly accessible kinases demonstrated structured activity patterns across perturbation contexts and exhibited interpretable signaling associations.

The findings suggest that protein-association evidence can effectively refine motif-derived KSR priors while maintaining broad kinome coverage. This framework offers a practical solution for more reliable phosphoproteomic kinase activity inference, ensuring that valuable information from dark kinases can be incorporated into the analysis.