In this study, we present a flexible framework to assess different combinations of computational algorithms and kinase-substrate libraries for the inference of kinase activities. For the benchmark, we use a set of kinase perturbation experiments to evaluate which combination is able to recapitulate the perturbed kinases from the phosphoproteomics data. Additionally, we propose a new benchmarking strategy based on multi-omics tumor data.
If you want to test your own method try out our package benchmarKIN and check out the documentation.
We have included the following kinase-substrate libraries:
- PhosphoSitePlus
- PTMsigDB
- Omnipath
- Gold Standard set of GPS 6.0
- iKiP-db
- NetworKIN
Additionally have tested the combination with predicted targets including the Kinase Library and Phosformer.
We have included the following methods for the comparison:
- fgsea
- Fisher's exact test
- KARP
- KSEA
- Kologomorov-Smirnov
- linear model - RoKAI
- Mann-Whitney-U test
- mean
- median
- multivatiate linear model - decoupler
- normalised mean
- Principal Component Analysis
- PTM-SEA
- sum
- univariate linear model - decoupler
- upper quantile
- VIPER
- z-score
- X-square test
Mueller-Dott, Sophia, Eric J. Jaehnig, Khoi Pham Munchic, Wen Jiang, Tomer M. Yaron-Barir, Sara R. Savage, Martin Garrido-Rodriguez, et al. 2024. “Comprehensive Evaluation of Phosphoproteomic-Based Kinase Activity Inference.” bioRxiv. https://doi.org/10.1101/2024.06.27.601117.