Figures
Defining biologically grounded validation standards for deep learning-based post-translational modification (PTM) prediction
The CLASPP framework combines unsupervised sequence-space stratification with supervised contrastive learning in protein language model embedding space to predict 12 PTM types from a single model. Training and experimental validation were performed in humans, while out-of-distribution testing spanned diverse model organisms across the evolutionary tree shown. The model’s utility in phosphoproteomics workflows is demonstrated through the annotation of ubiquitination sites in the understudied DCLK3 kinase. By integrating evolutionary breadth, experimental validation, and behavioral testing, CLASPP establishes a framework for rigorous benchmarking of next-generation multi-PTM prediction models (Gravel et al., 2026).
Image Credit: Nathan Gravel, licensed under CC BY 4.0
Citation: (2026) PLoS Computational Biology Issue Image | Vol. 22(8) September 2026. PLoS Comput Biol 22(8): ev22.i08. https://doi.org/10.1371/image.pcbi.v22.i08
Published: September 21, 2026
Copyright: © 2026 . This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
The CLASPP framework combines unsupervised sequence-space stratification with supervised contrastive learning in protein language model embedding space to predict 12 PTM types from a single model. Training and experimental validation were performed in humans, while out-of-distribution testing spanned diverse model organisms across the evolutionary tree shown. The model’s utility in phosphoproteomics workflows is demonstrated through the annotation of ubiquitination sites in the understudied DCLK3 kinase. By integrating evolutionary breadth, experimental validation, and behavioral testing, CLASPP establishes a framework for rigorous benchmarking of next-generation multi-PTM prediction models (Gravel et al., 2026).
Image Credit: Nathan Gravel, licensed under CC BY 4.0