RECIPE bridges transcriptomics and proteomics with deep graph learning on Ribo-seq data

Direct protein quantification remains incomplete, particularly in single-cell settings where current proteomics technologies capture only a small fraction of the proteome, limiting comprehensive characterization of protein-level regulation. Ribosome profiling provides a genome-wide readout of translation, yet converting these signals into accurate protein abundance estimates—especially for proteins lacking direct measurements—remains unresolved. Here we introduce RECIPE, a graph neural network framework that enables proteome-wide prediction of proteins beyond those directly measured by leveraging translation-resolved signals. RECIPE learns protein abundance from Ribo-seq–derived features (or transcriptomic features when Ribo-seq is unavailable) while using incomplete proteomics data for supervision, and integrates these signals with protein--protein interaction networks to capture context-aware relationships that improve inference for unmeasured proteins. Across human and mouse datasets, RECIPE enables proteome-wide prediction beyond experimentally measured proteins, recovering low-abundance and proteomics-undetected targets that are inaccessible to direct measurement, with particular advantages in data-sparse single-cell regimes. RECIPE provides a general framework for predicting protein abundance from translation-resolved (or transcriptomic) signals despite incomplete proteomics supervision, enabling proteome-wide analysis from incomplete molecular measurements.