Minimizing Fitters¶

Minimizing envelope scoring methods aim to minimize some “fit error” criterion, and discard solutions which don’t fit well. They tend to be quite sensitive and can be independent of the magnitude of the signal. This also means they do not handle detector noise or interference well. These methods can work well when targeting a list of compositions instead of exhaustively deconvolving an entire spectrum.

class ms_deisotope.scoring.GTestFitter(score_threshold=0.5)[source]¶

Evaluate an isotopic fit using a G-test

\[G = 2\sum_i^n{o_i * ({log}o_i - {log}e_i)}\]

where \(o_i\) is the intensity of the ith experimental peak and \(e_i\) is the intensity of the ith theoretical peak.

class ms_deisotope.scoring.ScaledGTestFitter¶

Evaluate an isotopic fit using a G-test after normalizing the list of experimental and theoretical peaks to both sum to 1.

\[G = 2\sum_i^n{o_i * ({log}o_i - {log}e_i)}\]

where \(o_i\) is the intensity of the ith experimental peak and \(e_i\) is the intensity of the ith theoretical peak.

class ms_deisotope.scoring.LeastSquaresFitter¶

Evaluate an isotopic fit using least squares coefficient of determination \(R^2\).

\[ \begin{align}\begin{aligned}{\hat e_i} &= e_i / max(e)\\{\hat t_i} &= t_i / max(t)\\{\hat t} &= \sum_i^n {\hat t_i}^2\\R^2 &= \frac{1}{{\hat t}}\sum_i^n ({\hat e_i} - {\hat t_i})^2\end{aligned}\end{align} \]

where \(e_i\) is the ith experimental peak intensity and \(t_i\) is the ith theoretical peak intensity

Maximizing Fitters¶

Maximizing envelope scoring methods aim to maximize some “goodness-of-fit” criterion, and discards solutions that don’t score highly enough. While not universally true, many of the maximizing scoring functions here are a function of the magnitude of the signal, which means that the threshold selected is signal magnitude dependent. This has the advantage of making the score threshold also a detector noise filter, but the threshold would now be instrument type-dependent.

A maximizing fitter with a well-chosen threshold is good at exhaustively deconvoluting a spectrum because it can more easily eliminate bad fits, but as isotopic pattern becomes more homogenous around the base peak of the isotopic pattern, they can more easily make mistakes identifying the monoisotopic peak. This is where MSDeconVFitter is less accurate than PenalizedMSDeconVFitter, albeit neither is perfect.

class ms_deisotope.scoring.MSDeconVFitter(minimum_score=10, mass_error_tolerance=0.02)¶

An implementation of the scoring function used in MSDeconV

\begin{split} s_{mz}(e, t) &= \begin{cases} 1 - \frac{\|mz(e) - mz(t)\|}{d} & \text{if } \|mz(e) - mz(t)\| < d,\\ 0 & \text{otherwise} \end{cases}\\ s_{int}(e, t) &= \begin{cases} 1 - \frac{int(t) - int(e)}{int(e)} & \text{if } int(e) < int(t) \text{ and } \frac{int(t) - int(e)}{int(e)} \le 1, \\ \sqrt{1 - \frac{int(e) - int(t)}{int(t)}} & \text{if } int(e) \ge int(t) \text{ and } \frac{int(e) - int(t)}{int(t)} \le 1,\\ 0 & \text{otherwise} \end{cases}\\ \text{S}(e, t) &= \sqrt{int(t)}\times s_{mz}(e,t) \times s_{int}(e, t) \end{split}

References

Liu, X., Inbar, Y., Dorrestein, P. C., Wynne, C., Edwards, N., Souda, P., … Pevzner, P. A. (2010). Deconvolution and database search of complex tandem mass spectra of intact proteins: a combinatorial approach. Molecular & Cellular Proteomics : MCP, 9(12), 2772–2782. https://doi.org/10.1074/mcp.M110.002766

class ms_deisotope.scoring.PenalizedMSDeconVFitter(minimum_score=10, penalty_factor=1, mass_error_tolerance=0.02)¶

An Isotopic Fitter which uses the MSDeconVFitter score weighted by 1 - penalty_factor * ScaledGTestFitter score

\[S(e, t) = M(e, t) * (1 - G(e, t))\]

where \(e\) is the experimental peak list and \(t\) is the theoretical peak list

Other Fitters¶

class ms_deisotope.scoring.FunctionScorer(function, minimum_score=10, selector_type=MaximizeFitSelector)¶

Use a user-provided Python function or callable object to evaluate an isotopic envelope fit.

The provided function should take two arguments, a list of experimental fitted peaks and a list of theoretical peaks, and return a single number as an output.

Make sure to pass the right thresholds and selector types when creating an instance of this class that match the semantics of the provided function.

While this type is available for convenience, there is considerable overhead in using it compared to one of the C accelerated fitters.