PSMStatistics

PeptideSpectrumMatching reports several scores for every spectrum (SEQUEST-like, Andromeda-like, X!Tandem-like, plus mass error, peptide length and how far the best candidate is ahead of the next one). PSMStatistics learns one combined score from these columns. It uses the target and decoy labels of the matches as training signal, retrains in iterations while the set of confident targets grows (semi supervised), and stops when an iteration adds too few new positives or the iteration limit is reached. From the combined score it computes a q-value and a PEP value per match and keeps the PSMs under both thresholds.

PSMStatistics also reads the output of PeptideSpectrumMatchingTIMs. Both tools write the 28 column layout of this release, and .psm files written by earlier releases (23 columns) have to be regenerated before this release of PSMStatistics reads them. When every row carries finite Hyperscore and Expectscore values, it uses both as model features, with Expectscore represented by its negative log10, and seeds the first training round with the expectation value. Without them the SEQUEST-like score seeds it, as before.

Inputs and outputs

For every input run.psm the tool writes run.qpsm to the output directory. The file is tab separated with a header and holds one row per scan: the identifiers of the peptide and its modification state, the search scores, the combined ModelScore, QValue, PEPValue, the sequence and the protein names. ProteinInference and PSMBasedQuantification read .qpsm files.

The tool also creates a run_plots directory per input. It stays empty unless you pass -dc, which adds Metrics.html and InitialSeparation.html, plus one separationAtIteration_<n>.html per training iteration. Charts exist only for the estimated threshold. Log output goes to PSMStatistics_log.txt and run_log.txt in the output directory.

Parameters

The fields of Threshold.Estimate:

The default file is pSMStatisticsParams.json.

Writing a parameter file

open ProteomIQon
open ProteomIQon.Domain

let psmStatisticsParams : Dto.PSMStatisticsParams =
    {
        Threshold =
            Threshold.Estimate
                {
                    QValueThreshold                  = 0.01
                    PepValueThreshold                = 0.05
                    MaxIterations                    = 15
                    MinimumIncreaseBetweenIterations = 0.005
                    PepValueFittingMethod            = PepValueFittingMethod.IRLS
                }
        ParseProteinIDRegexPattern = "id"
        KeepTemporaryFiles         = true
    }

// Replace the temp folder with your project folder.
let outputPath = System.IO.Path.Combine(System.IO.Path.GetTempPath(), "pSMStatisticsParams.json")

Json.serializeAndWrite outputPath psmStatisticsParams

Outputs

The resulting .qpsm file is generated as a tab-delimited text file. These columns can be inspected individually to examine specific search results, scoring metrics, and identification parameters.

Running the tool

Install with dotnet tool install --global ProteomIQon.PSMStatistics, then score one run:

proteomiqon-psmstatistics -i path/to/run.psm -d path/to/database.db -o path/to/output -p path/to/pSMStatisticsParams.json

Several runs at once, three of them in parallel:

proteomiqon-psmstatistics -i path/to/run1.psm path/to/run2.psm path/to/run3.psm -d path/to/database.db -o path/to/output -p path/to/pSMStatisticsParams.json -c 3

Add -dc to write the separation charts for each iteration:

proteomiqon-psmstatistics -i path/to/psmFolder -d path/to/database.db -o path/to/output -p path/to/pSMStatisticsParams.json -c 3 -dc

All flags:

proteomiqon-psmstatistics --help