Implementing Annotation Confidence Scoring in Untargeted Mass Spectrometry Workflows for Small Molecule Analysis

Untargeted small molecule analysis by high-resolution
mass spectrometry
is integral to environmental and biological research, enabling comprehensive
characterization of complex samples. However, data interpretation
and reporting remain challenging due to the complexity and high dimensionality
of molecular features in untargeted data sets. Current data analysis
platforms provide integrated tools for processing and annotation yet
lack a standardized framework for assigning and reporting annotation
confidence. Communicating varying levels of confidence in untargeted
data sets continues to pose challenges without an automated ranking
system. To address this, a custom scripting node was developed that
assigns annotation confidence levels based on the widely adopted Schymanski
et al. scoring scheme. While implemented here for metabolomics, the
scoring approach is broadly applicable to other untargeted small molecule
workflows. The script can be incorporated into a commercial data analysis
software package and functions as a standalone postprocessing node,
expanding the original five-level system with four new sublevels (levels
3a/3b and 4a/4b) to improve specificity and distinguish cases that
fall between established categories. Annotation confidence is assessed
using available information from all compound identification workflow
nodes (e.g., Predicted Composition, mzVault, mzCloud, ChemSpider),
and the consensus scoring algorithm is used to evaluate agreement
among search nodes for greater accuracy. Validation using NIST SRM
1950 plasma samples demonstrated 100% scoring accuracy in negative
mode and >99.5% in positive mode across both RPLC and HILIC separations.
This tool enhances data reporting, improves transparency, and promotes
consistency across studies, facilitating standardization and comparability
of untargeted metabolomics results.