Match-between-runs feature grouping via the PIP-ECHO algorithm.
Groups features that represent the same identified peptide across multiple retention-time-aligned LC-MS runs of a single fraction. Identifications are transferred to runs in which a feature was detected but not identified (match-between-runs); candidate transfers are scored and filtered at a controlled false-discovery rate using a target/decoy model.
Parameters of this class are:
| Name | Type | Default | Restrictions | Description |
|---|
| fdr |
float | 0.05 |
min: 0.0 max: 1.0 | MBR FDR threshold (0.05=5%). |
| random_seed |
int | 0 |
| Seed for the random number generator used to select decoy donors. A fixed seed makes results reproducible. |
| min_decoys |
int | 20 |
min: 1 | Minimum number of MBR decoy transfers required before any transferred feature is kept. Transferred features are also dropped whenever the requested 'fdr' cannot be resolved by the available decoys (1/decoys > fdr). When transfers are dropped, only direct identifications are retained. A 'fdr' of 1.0 disables FDR control and keeps all transfers regardless of this value. |
| max_training_points |
int | 50000 |
min: 0 | Upper bound on the number of candidate transfers used to TRAIN the transfer-FDR SVM in each cross-validation fold (0 = unlimited). Predictions and the FDR/q-values are always computed over ALL transfers, so this only bounds the SVM-fitting cost and does not change which transfers are scored. Runs with very many features per run can otherwise make the SVM grid search slow. When the cap is hit, a deterministic stratified subsample is used: both labelled classes (decoys and positives) are kept whole if they fit, otherwise the minority class is kept whole and the majority is score-spread sampled to fit the budget. |
| distance_RT:max_difference |
float | 100.0 |
min: 0.0 | Never pair features with a larger RT distance (in seconds). |
| distance_MZ:max_difference |
float | 10.0 |
min: 0.0 | Never pair features with larger m/z distance (unit defined by 'unit') |
| distance_MZ:unit |
string | ppm |
Da, ppm | Unit of the 'max_difference' parameter |
| local_rt:enabled |
string | true |
true, false | Use a LOCAL adaptive retention-time window for MBR candidate search instead of the single global RT window ('distance_RT:max_difference'). For each donor the expected acceptor RT is predicted from nearby peptides identified in BOTH runs (local alignment) and the search window is sized from the local RT scatter, sharpening the RT feature and removing much of the false in-window background. The window is widened adaptively if too few decoys are produced to resolve the FDR. Set 'false' to restore the legacy single global RT window. |
| local_rt:max_window |
float | 100.0 |
min: 0.0 | Maximum half-width (seconds) of the local RT window before adaptive widening (backstop cap on the data-driven width). |
| local_rt:min_window |
float | 5.0 |
min: 0.0 | Minimum half-width (seconds) of the local RT window. |
| local_rt:anchor_window |
float | 120.0 |
min: 0.0 | Only peptides within this RT distance (seconds) of the donor are used as local-alignment anchors. |
| local_rt:anchors |
int | 3 |
min: 1 | Maximum number of anchor peptides per side used for the local RT prediction. |
| local_rt:sigma_scale |
float | 3.0 |
min: 0.0 | Local RT window half-width = this multiple of the local anchor RT-shift standard deviation. |
| local_rt:fallback_window |
float | 15.0 |
min: 0.0 | Half-width (seconds) used when fewer than two local anchors are available. |
| local_rt:auto |
string | true |
true, false | Estimate the local RT window scales (max_window, min_window, anchor_window, fallback_window) from the data instead of using the fixed values above, so one configuration adapts across short and long gradients. Derived from the RT-shift scatter of shared anchors and the anchor density; uses 'median_fwhm' as a physical lower guard when provided. |
| local_rt:median_fwhm |
float | 0.0 |
min: 0.0 | Chromatographic peak FWHM (seconds) used as a physical lower guard when 'auto' is enabled (0 = estimate window scales from RT residuals and anchor density only). Set by the host tool (e.g. ProteomicsLFQ) which has raw-spectra context. |
| local_rt:rt_score |
string | svm_and_mbr |
raw, svm, svm_and_mbr | How the retention-time feature is used on the local RT path. 'raw': the raw |Δrt| is the SVM predictor (legacy). 'svm': a calibrated [0,1] RT-agreement score (FlashLFQ-style two-tailed CDF against the data-driven RT prediction-error distribution) replaces it as the SVM predictor. 'svm_and_mbr': as 'svm', and the calibrated score also enters the bootstrap geometric mean. Ignored unless 'enabled'. |
Note:
- If a section name is documented, the documentation is displayed as tooltip.
- Advanced parameter names are italic.
Public Member Functions inherited from FeatureGroupingAlgorithm
1.9.8 