The ReSpect algorithm accepts a significant false positive rate to avoid the risk of false negatives. For this reason, the application applies a scoring algorithm to the ReSpect results to determine which components are the most reliable. This scoring algorithm reduces the false positive rate by calculating a Bayesian fitness measure that distinguishes reliable components from components that could represent harmonics, fortuitous fits to noise, or both. The algorithm compares the probability that a predicted component might produce peaks with an observed set of m/z positions and an intensity profile to the probability that this particular pattern could occur by chance. It then applies an additional set of penalties to exclude low-intensity components, harmonics, and superfluous fits to successive charge states of successive valid components.

The score calculation algorithm does not require you to set any control parameters. It copies any parameters that it might need from the corresponding control parameters used by the ReSpect algorithm or derives them from the ReSpect results themselves. However, you can use the Quality Score Threshold parameter to set a threshold value and exclude components with scores below the threshold value. See ReSpect deconvolution parameters.

Deconvolution scores are absolute rather than relative so that you can compare scores between spectra, as well as compare scores within an individual spectrum. The higher the score, the better the component quality is. "Better" components in one spectrum have higher scores than "worse" ones in another spectrum. For example, a score of 100 in one spectrum represents a more reliable fit than a score of 50 in another spectrum. In general, components with higher molecular weights, higher intensities, or both that might tend to have more charge states, receive higher scores.

For high-quality spectra, you might want to ignore components with scores below 40. For lower-quality spectra, this threshold might be 10 or lower. In general, scores indicate the following:

  • In almost all cases, scores below 0 signify "bad" components.
  • In particular, scores significantly below –100 are associated with harmonics and other unreliable fits that you should ignore.
  • Scores between 0 and 20 indicate components that might be questionable.
  • The precise value of this threshold depends on the quality of the spectra themselves. For clean spectra with well-defined and widely separated components, all results with a score greater than 0 might be good. As the noise increases and the spacing between components decreases, this threshold might rise. The precise threshold below which you might want to ignore components depends on the quality of the spectrum, the peak confidence level, and the mass tolerance that the ReSpect algorithm uses. Tests indicate that 20 might be a good starting point when you select a threshold.
  • Scores between 20 and 40 generally indicate "good" components.
  • Scores above 40 indicate components of very high quality.

After the deconvolution processing of the spectrum, you can view the quality score for each component in your analysis in the Score column of the Results table on the Process and Review page. Sort the scores from high to low by selecting the down arrow next to the Score column header, or sort the scores low to high by selecting the up arrow.