Processing by non-target screening

Introduction

Measurement files are processed by two algorithms in NTSPortal, library screening (dbas) and non-target screening (nts), the later is described here. The results of both processing algorithms are stored in feature tables in Elasticsearch. Processing by nts only includes features without a compound match in the CSL (i.e. unknowns). To achieve this, nts processing includes an annotation step, but annotated features (i.e. with a match in the CSL) are removed from the final feature list (during the conversion to featureRecord step). This removal does not include all possible adducts, in-source fragments or isotopogues of a compound, since the CSL can not include all possible permutations. Therefore, there may still be some duplication in the results.

Processing by nts works analogously to dbas and can be started with the same various screening*() functions using nts as the screeningType argument. The rest is the same as dbas: Results are converted to a list of featureRecords and saved as .RDS files. These are imported into the database using ingestFeatureRecords().

Workflow details

File scanning

Files are processed together in batches. The processing algorithm using functions provided by ntsworkflow and performs the following steps:

1. Peak-picking: An initial feature list is built by searching EICs for chromatographic peaks
2. Feature alignment: Features are grouped across samples (m/z and $t_R$ similarity), building the alignment table
3. Alignment table cleaning: The alignment table is filtered by different means, e.g. only keeping features found repeatedly in replicate injections.
4. Annotation with the CSL: Using m/z, $t_R$ and MS² matching with with collective spectral library
5. Internal standards marked: The IDs of internal standards are recorded in the sample list (for building the featureRecord).
6. Blank correction: Features found in samples and field blanks (background) are removed from the alignment table. Internal standards are included in the field blank so they are also removed from the alignment table in this step (however, they remain in the peak list).

The result of this processing is a ntsResult object, a list containing 4 tables (tbl_df objects).

This is a test caption

Name of table	Comment
peakList	Peaks (features) detected
sampleList	Table of measurement files and associated metadata
alignmentTable	Table of aligned features
annotationTable	Table of annotations of the alignmentTable (matches with the CSL)

Figure: NTS file scanning steps

Conversion to featureRecord

The ntsResult object is converted to a featureRecord using a method of the convertToRecord() generic. It includes the following steps:

1. Annotated rows of the alignmentTable (those features with a match in the CSL) are removed.
2. Blank files and all associated features and data are removed.
3. Peak areas of all features are collected from the alignmentTable.
4. Peak areas of internal standard features are gathered from the peakList. These were removed from the alignment table during step 7 of file scanning (but remained in the peak list table). The IDs were recorded in step 6 of file scanning.
5. featureRecords are enriched with EICs, MS1 and MS2 spectra, which are collected from the raw measurement files.
6. The list of featureRecords is saved as a .RDS file.

Figure: Conversion of ntsResult to a featureRecord

Each featureRecord also includes the path to the measurement file (to reference the msrawfiles entry and all the associated metadata) and the feature table alias, which is used to create the feature table in Elasticsearch.

The featureRecords are collected in a list and saved as one RDS file per batch.