We have recently shown that using fragment ion intensities can drastically increase the confidence in peptide identification and reduce false positive matches compared to classic database searching. Further peptide properties such as precursor charge and the rather vague term ‘proteotypicity’, influence the detectability of peptides. With the advent of deep learning and the wealth of data stored in public repositories, one may eventually be able to predict all peptide properties that are relevant for the detection and identification of peptides. Here we report on the extension of Prosit, a deep learning architecture for peptide property prediction, to deal with non-tryptic and multiply modified peptides as well as predicting proteotypicity and precursor charge.