Fragment ion intensity prediction boosts TIMSTOF peptide ID rates!

 

This was on here as a preprint, I think, but I also think it got better during peer review.

The idea is pretty simple. The thing about the millions of spectra that things like PROSIT use for deep learning is that they were generated on Orbitraps. Makes sense. Super high quality spectra. However, fragmentation energies and energy ramps and mass analyzer architectures are all different between this and other instruments. Heck, it's probably not out of the question to think that the Orbitrap-TOF Asstral fragmentation patterns are probably not perfectly matched to Orbitrap spectra. Maybe they are, but there are definitely differences between other vendor TOFs and these library spectra.

This group works through those differences and finds they can drastically (3-fold more!) improve identification rates by retraining the models!