FastSpel: A Method for Fast Spectral Library Generation

Recently, several methods have been proposed for predicting
peptide
MS/MS fragment intensity profiles. These predicted profiles may be
used for generating spectral libraries for data-independent acquisition
analysis, or for improving peptide identification by rescoring the
peptide-spectra matches identified by search engines such as MaxQuant.
Although some of the proposed intensity prediction methods generate
high quality spectral libraries and significantly improve peptide
identification, they are computationally expensive and their parameters
are difficult to interpret. In this paper, we introduce FastSpel (fast spectral library), a fast and interpretable fragment intensity prediction method
for tryptic peptides. Testing FastSpel on 23 independent data sets,
we show that its performance, in terms of improving peptide identification
via rescoring and spectral library generation, is comparable with
those of the existing state-of-the-art methods, while being over 2
orders of magnitude less computationally expensive than these methods.
Moreover, analysis of parameters of the model corroborates known fragmentation
rules, such as the “proline effect”, and suggests novel
patterns. In addition to FastSpel, we propose a simple scoring function
that achieves rescoring/identification performance close to that of
Percolator, a widely used program for this purpose, without requiring
model training as Percolator does.