Post-translational modifications (PTMs) are essential to the correct functioning of proteins. While chromatography coupled to tandem mass spectrometry (LC-MS/MS) allows for a high-throughput investigation of PTMs, bioinformatics tools to analyze the resulting data are lacking. Indeed, due to the large variability in potential PTMs or artefactual modifications, proteomics search engines suffer from a decreased identification sensitivity when including modifications into the search space. Deep learning models that accurately predict peptide LC-MS/MS behavior have shown to recover the identification sensitivity, although such predictors are limited to unmodified peptides. Very recent work by me and my colleagues on peptide retention time prediction has shown that deep learning models can generalize predictions across modified and unmodified residues. I therefore here propose to extend these methods to the prediction of peptide MS/MS spectra, enabling these spectra to be predicted for any peptide modification. Furthermore, by integrating the two prediction tasks and investigating more advanced deep learning techniques, I aim to push prediction accuracy to the limit of technical variance. I will also leverage these novel predictors for much increased identification sensitivity, and for improved localization of PTMs. Finally, I will apply these novel tools on a large amount of public proteomics data to generate the most detailed view ever of the proteome-wide PTM landscape.