Predicting the best possible quantitative peptides for a protein is still not a well-solved problem. One of the early papers which Skyline was in part designed to support describes an entirely empirical method of assessing this with expressed proteins. (Stergachis, et al. Nature Methods 2011 - https://pubmed.ncbi.nlm.nih.gov/22056677/) It details how poorly existing predictors were functioning at the time for this operation. One of the problems being that "proteotypicity" itself was originally conceived as the probability of a peptide being detected in MS/MS from a DDA run, possibly even by simply counting the number of such identified MS/MS spectra in a large public library like PeptideAtlas. Mike MacCoss has argued in the past that this is a terrible proxy for suitability for targeted quantification, because whether a peptide precursor feature is sampled with MS/MS in a DDA run depends highly on what other features are eluting at the same time, and interference in MS1 can produce chimeric MS/MS spectra which are less likely to be identified reliably.
Brian Searle while in the MacCoss lab took a crack at improving upon those existing methods using DIA data, and the result was published (Searle, et al. MCP 2015) and is now in the Skyline tool store as the tool Prego:
You can install it into Skyline using the Tools > Tool Store menu item.
The Prosit team is also working on a new predictor for this, which we will try to integrate into Skyline with the other Prosit support already available in Skyline.
If you have other suggestions of the tools that have worked best for you in the past, please add that to your feedback. Otherwise, know that we do watch this space and Mike MacCoss has encourage his students to explore it and try to improve the field's understanding of how best to approach choosing peptides for reliable quantitative mass spectrometry.
Thanks for posting your feedback to the Skyline support board.