#33 · Proteomics

Prior-knowledge networks and phosphosite models refine proteomic inference

The dominant thread is the use of prior knowledge, statistical models and learned sequence models to interpret proteomic and phosphoproteomic measurements. MSstatsBioNet connects MSstats differential protein abundance results to INDRA literature-derived and curated biomolecular networks, while a separate multi-omics study extends DrDimont networks with kinase interactions and applies them to pulmonary arterial hypertension endothelial cells, highlighting MAPK signalling. A pan-cancer analysis of 1,006 tumours uses nested linear regression to link 3,038 phosphosites to the abundance of interacting proteins. On the sequencing side, InstaNovo-P adapts transformer-based de novo peptide sequencing to phosphoproteomics to improve phosphosite localization. LiMeNEx applies comparable network logic to lipid biochemistry, linking lipids, enzymes and transcription factors across 504 curated reactions. What is new is the joint use of curated knowledge networks with quantitative measurements, with several items delivering accessible resources such as a Bioconductor package and webservers. Because the items span different modalities, the shared element is integration rather than a single method.