This paper proposes a fuzzy support vector machine (SVM)-based data fusion technique that aims to distinguish true or false peptide-spectrum matchings (PSM) from integrated SEQUEST scores. The proposed system, MFS, has comparable performance with other existing systems (PeptideProphet and Percolator) that perform the same task, according to the experiments performed by the authors. However, this system considerably reduces the time required to run this task.

PeptideProphet employs an empirical Bayesian approach for model fitting, utilizing a pre-trained model created with a collection of true and false peptide identifications. Percolator, on the other hand, can learn the unique feature of each dataset, hence improving the accuracy with an SVM-based method.

The strongest part of the paper, what makes this method faster, is the fuzzy membership approach; the authors use it to alleviate the impact of noise and eliminate the effect of untrustworthy labels in the dataset.

Although the authors claim that their system runs faster than the existing systems, they do not provide the comparison results regarding timings of these three systems; this is the weakest part of the paper. There is no discussion about what makes their system faster. Another weak point is that, although the results MFS provides have a very high overlap with PeptideProphet and Percolator, it identifies fewer target PSMs, especially compared to Percolator.

I would recommend reading this paper along with PeptideProphet and Percolator publications in order to better understand how these systems work and judge the advantages and disadvantages of each approach.