A Framework for Mass Spectral Quality Assessment without Prior Information

It is well known that a majority of experimental spectra are of too poor quality to be interpreted by any automatic method. It wastes time to interpret these "un-interpretable" spectra. On the other hand, some spectra with high quality also cannot be interpreted by any automatic method, but maybe by manual checking. Therefore, it is worthwhile to develop a powerful filter that could eliminate those spectra with poor quality before any interpretation. This paper proposes a framework to assess the quality of tandem mass spectra without prior information. The proposed framework includes: (1) filtering noises from the experimental mass spectra; (2) extracting the peaks; (3) mapping each spectrum into a feature vector which describes the quality of experimental spectra; (4) classifying spectra into clusters by using an unsupervised classification method; (5) training earning a classifier using the cluster with the high quality spectra and the one with poor quality spectra; and (6) assessing all spectra by using the trained classifier. The proposed framework has been implemented and tested on two tandem mass spectra datasets acquired by ion trap mass spectrometers. Computational experiments illustrate that the method based on the proposed framework can eliminate majority of poor quality spectra while losing very minority of high quality spectra.