 |
OpenMS
|
Running summary of the complete candidate pool of one spectrum. More...
#include <OpenMS/ANALYSIS/ID/ProSEAlgorithm.h>
Public Member Functions | |
| void | add (double score) |
| Fold one freshly scored candidate into the summary. | |
| double | deltaScore () const |
| Best minus runner-up over the full pool. | |
| double | zScore () const |
| How much of an outlier the best score is within its own candidate pool. | |
Public Attributes | |
| double | sum = 0.0 |
| sum of all candidate scores | |
| double | sumsq = 0.0 |
| sum of squared candidate scores | |
| double | best = 0.0 |
| best candidate score (HyperScore is non-negative, so 0 doubles as "none seen") | |
| double | second_best = 0.0 |
| runner-up candidate score | |
| Size | count = 0 |
| number of candidates scored | |
Running summary of the complete candidate pool of one spectrum.
scoreSpectraAgainstIndex_() prunes each spectrum to max(report_top_hits_, 2) candidates as soon as it has scored them, so by the time postProcessHits_() runs the surviving hits are no longer a sample of the search space: with the default report:top_hits=1 only two candidates remain, and derived features such as ln_num_candidates or hyperscore_zscore would degenerate into a binary flag and a rescaled delta score respectively.
add() is therefore called for every candidate the moment it is scored – before pruning and before zero-scoring candidates are dropped – so the summary reflects the full pool. Instances accumulate across chunks in the chunked search paths, where each chunk contributes its own candidates for the same spectrum.
|
inline |
Fold one freshly scored candidate into the summary.
|
inline |
Best minus runner-up over the full pool.
| double zScore | ( | ) | const |
How much of an outlier the best score is within its own candidate pool.
Standard score of best against the mean and (population) SD of all count candidates – a lightweight significance proxy, since HyperScore (unlike e.g. MS-GF+'s SpecEValue) has no closed-form e-value.
Standardising against the whole pool rather than against the count - 1 non-best candidates keeps the statistic bounded: by Samuelson's inequality a pool member cannot deviate from its own mean by more than sqrt(count - 1) SDs, so the feature tops out at sqrt(scoring:max_candidates_per_spectrum - 1) (7 at the default cap of 50). The leave-one-out form has no such bound – with two candidates the SD of the single remaining score is 0 by construction, and the ratio diverges. That mattered in practice: on a 70k-PSM Orbitrap run the leave-one-out form produced values up to 6.5e+07, which is enough to dominate Percolator's feature standardisation.
Returns 0 when fewer than two candidates were scored, and when every candidate scored the same – in both cases the best candidate does not stand out from the pool.
| double best = 0.0 |
best candidate score (HyperScore is non-negative, so 0 doubles as "none seen")
| Size count = 0 |
number of candidates scored
| double second_best = 0.0 |
runner-up candidate score
| double sum = 0.0 |
sum of all candidate scores
| double sumsq = 0.0 |
sum of squared candidate scores
1.9.8 