In this work, we study the problem of protein-isoform inference from mass spectrometry proteomics data. Inferring protein isoforms is complex because proteins are only indirectly measured via peptides. Here, we propose a Bayesian approach, based on a two-layer latent variable model to recover protein isoforms, starting from peptide-level data. Since isoform-level data is scarse, we further enhance information by embedding transcriptomics data (i.e., mRNA abundance), which are incorporated via informative priors. Our approach allows inferring the presence/absence and abundance of individual protein isoforms, and provides a measure of uncertainty of both estimates, via posterior probabilities and highest posterior density credible intervals. Notably, at present, existing proteomics tools do not allow inferring protein isoform abundances. Here, we present preliminary results, based on simulation studies, while more extensive benchmarks on real data are currently being performed. Our framework may be valuable for life scientists, and enable them to gain deeper insight into key biological mechanisms.