That's a pretty broad question. I am by no means an expert (and I hope someone who is will weigh in here), but obviously it depends on how well the algorithm models the actual performance characteristics of your mass spec. That's going to depend on a lot of factors, not least of which is what kind of molecules you're interested in. Some are easier than others to model in fragmentation.

Certainly there have been huge steps made with machine learning recently, but I'd think you'd at least want to try it both ways with a representative set of molecules and see how well the two library versions agree before going all in on in silico.

Brian

Hi Mehul,
Almost all of the tools, DIA-NN, Spectronaut, EncyclopeDIA, FragPipe, etc... approach the analysis by first shrinking the list of peptides to those that are detectable in the sample. Historically, this was done using DDA and fractionation. However, now tools like Spectronaut and FragPipe take a spectrum-centric approach to find peptides and then build a "library" to be used either with an mProphet algorithm or DIA-NN. DIA-NN does a two-step search as well where it will search all or a subset of the files to build a chromatogram library and then use that reduced library to analyze the individual samples. We have always advocated for using a few runs that incorporate gas-phase fractionation into the acquisition to improve the sensitivity and specificity to aide in the first step but it is totally possible to just use a subset or all of your files for this. This strategy works well for FragPipe, EncyclopeDIA, or DIA-NN. See these papers Pino et al, MCP, Searle et al, Nat Comms 2018, and Searle et al, Nat Comms, 2020.

In Skyline it is possible to search with EncyclopeDIA using a few runs to build a chromatogram library (either gas-phase fractionated or not) and then use that library to analyze the files individually. You can also use DIA-Umpire and a spectrum-centric search of the data and then use that library to use the advanced mProphet models within Skyline. MS-Fragger-DIA can also be run inside of Skyline and then use the mProphet scoring of those identified peptides. You can also run FragPipe outside of Skyline and create a Skyline document directly. Also run DIA-NN and then import the detected peptides and search results.

Skyline is very flexible in the way that it lets you analyze data with different tools both within and outside of Skyline. All of this comes down to multiple hypothesis testing. You need a strategy to reduce the list of peptides otherwise you will struggle with sensitivity.

Cheers,
Mike