AlignmentBasedQuantStatistics

A peptide ion transferred by alignment and quantified by AlignmentBasedQuantification has no identification behind it, so nothing says yet whether the peak found at the predicted scan time is the peptide. AlignmentBasedQuantStatistics gives every transferred quantification an alignment score and an alignment q-value, so downstream tools can filter transferred quantifications the way identifications are filtered by q-value. It learns from peptide ions where the truth is known: the .align file also predicts ions the run had identified itself, and for those the transferred quantification can be compared with the quantification of the identified ion. Ions where m/z and intensity agree closely are positive training examples, ions where they disagree are negative ones. A gradient boosted tree classifier trained on them scores every transferred ion, and the q-value follows from the score distribution of the negative examples.

Inputs and outputs

-i to -iii name the runs to score, -l to -lll the runs to learn from. The runs to learn from can be the same runs. Every one of the six flags takes one file, several files, or a directory that is searched for *.quant or *.align. Without -mf the quant, align and aligned quant files are paired by position. With -mf they are paired by file name without extension. Every scored run trains a classifier on the pooled examples of all training runs.

The tool writes a file with the same name as the -i .quant file into the output directory. It holds the rows of that file with AlignmentScore and AlignmentQValue filled for the rows whose QuantificationSource is Alignment. JoinQuantPepIonsWithProteins reads it, and LabelFreeProteinQuantification and LabeledProteinQuantification filter on the q-value column through their Alignment_QValue parameter. With -ts the tool also writes <quant file name>testset, the training examples taken from the run of the same name with their aligned quantification columns, AlignmentScore, AlignmentQValue and the label column WasPositiveSet. With -dc it writes ProbabilityHistogram.html and QValueDistribution.html into a directory named after the run. Logs go to AlignmentBasedQuantStatistics_log.txt and one <run>_log.txt per scored run in the output directory.

Parameters

The four cutoffs decide which training ions count as positive and which as negative. Each compares a transferred value with the value of the identified ion (light or heavy, matching the ion's GlobalMod) and expresses the allowed difference as a fraction of the identified value.

The positive test runs first. An ion that fails it is negative, whether or not it passes one of the negative conditions. The default file is AlignmentBasedQuantStatisticsParams.json.

Writing a parameter file

open ProteomIQon

let alignmentBasedQuantStatisticsParams : Dto.AlignmentBasedQuantStatisticsParams =
    {
        PositiveQuantMzCutoff = 0.01
        NegativeQuantMzCutoff = 0.99
        PositiveQuantCutoff   = 0.1
        NegativeQuantCutoff   = 0.9
    }

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

Json.serializeAndWrite outputPath alignmentBasedQuantStatisticsParams

Running the tool

Install the tool with dotnet tool install --global ProteomIQon.AlignmentBasedQuantStatistics. Score one run and learn from the same run.

proteomiqon-alignmentbasedquantstatistics -i path/to/run.quant -ii path/to/run.align -iii path/to/aligned/run.quant -l path/to/run.quant -ll path/to/run.align -lll path/to/aligned/run.quant -o path/to/output -p path/to/AlignmentBasedQuantStatisticsParams.json

Score every run in the directories, learn from all of them, pair the files by name, and work on four runs at a time.

proteomiqon-alignmentbasedquantstatistics -i path/to/quant -ii path/to/align -iii path/to/aligned -l path/to/quant -ll path/to/align -lll path/to/aligned -o path/to/output -p path/to/AlignmentBasedQuantStatisticsParams.json -mf -c 4

Add -ts to keep the training examples and -dc to get the charts.

proteomiqon-alignmentbasedquantstatistics -i path/to/quant -ii path/to/align -iii path/to/aligned -l path/to/quant -ll path/to/align -lll path/to/aligned -o path/to/output -p path/to/AlignmentBasedQuantStatisticsParams.json -mf -ts -dc

Print the description of every argument.

proteomiqon-alignmentbasedquantstatistics --help