T2Pdecoder enables protein-centric analyses from transcriptomic data.

Protein quantification is not as extensive as RNA quantification, especially for isocitrate dehydrogenase (IDH) mutant gliomas. Predicting protein abundance from RNA is valuable for leveraging existing data to understand biological processes, though the weak correlation between RNA and protein poses a significant challenge. Most existing methods predict limited protein subsets from transcriptome, constraining their broader proteomic applications. Here, we present T2Pdecoder, an integrative multi-omics deep learning model designed to predict broad protein abundance profiles by learning the shared embedding space of protein and RNA. T2Pdecoder is evaluated on different glioma datasets, achieving modest but consistent improvements over RNA-only baselines in concordance with measured protein abundance, while more accurately recapitulating protein-level pathway enrichment patterns. The applications of T2Pdecoder on glioma bulk RNA data uncover functional subgroups with significant survival differences. Furthermore, T2Pdecoder reduces batch-associated variation in single-cell RNA data and identifies distinctive cell markers. Collectively, these results suggest that T2Pdecoder enables protein-centric analyses from transcriptomic data and may provide complementary biological insights beyond conventional RNA-only analyses in cancer research.