Will Thompson   Will Thompson, Ph.D. its currently Principal Scientist in Research and Development at 908 Devices Inc, and an Adjunct Assistant Research Professor at Duke University. Prior to his current role at 908 Devices, Will spent 15 years in the Duke Proteomics and Metabolomics Shared Resource, where his team established the metabolomics arm of the core starting in 2013. Will has been an active Skyline user, zealot, and requester-of-features since 2009.

Enhancing Skyline for MicroChip CE-HRMS Metabolomics with a Novel Software Pipeline Including Automated System Suitability Testing, Data QC, and Metabolite Peak Quantification

Introduction
Metabolomics has demonstrated the ability to measure hundreds of metabolites in diverse sample types. Nonetheless, data analysis remains a key bottleneck in targeted and nontargeted approaches. For example, although system suitability testing (SST) has been widely adopted as common practice in metabolomics, human intervention is often required to accept or reject system suitability, without strict go/no go criteria. After experiments have been run, data quality assessment is typically performed manually and many hours after experiment completion, resulting in time wasted if a sample needs to be reanalyzed. Finally, errors in metabolite/peak assignment require arduous manual curation. We have developed a novel Windows application which automates instrument orchestration, SST interrogation, raw data QC, and the quantitative pipeline for ZipChip CE-MS metabolomics.
Read More
Methods
Amino acid standards (Promega) were diluted in Peptides Diluent (908 Devices) and utilized as a system suitability standard (SST) for ZipChip CE-MS analysis on an Exploris 240 (Thermo). Pass/Fail criteria were developed using historical criteria from hundreds of analyses, and included migration time, peak shape (resolution), and migration index. Samples including plasma, serum, and urine (BioIVT) were extracted using a simple methanolic protein precipitation including 36 stable isotope labeled internal standards and ammonium acetate. Open source application was developed in Python with standard data science toolset (e.g. pandas, numpy). The application utilizes Skyline tempas a backend for data extraction and for steps of the quantitation pipeline.
Preliminary Data
We determined that a set of 5 metabolites in a Skyline template was sufficient to detect system failures nearly perfectly in CE-MS. Using a set of several hundred analyses across multiple chips, we established pass/fail criteria for the following metrics in SST: 1) migration time of lysine (highest mobility), 2) migration time of aspartic acid (lowest mobility), 3) separation resolution of isoleucine and leucine, and 4) migration of valine compared to Ile/Leu. These metrics were then incorporated into a GUI which provides users immediate feedback at system startup, with guided procedures for next steps if the SST fails.
To automatically detect system/analysis failures during a batch, we screened 36 internal standards and identified 4 measurements that could detect in an automated manner whether the sample was injected properly, if the sample was prepared correctly, and whether the separation quality was acceptable. Using several hundred runs to build acceptance criteria from a variety of sample types (serum, plasma, urine) we utilized a Skyline template against four internal standards to determine key failure modes with nearly 100% accuracy, allowing automated resubmission or exclusion from data analysis as required.
Finally, we utilize the iRT calculator approach to index migration time (iMT) for a set of 325 metabolites, and store this iMT library for subsequent annotation of peaks. We demonstrate that despite accurate time predictions, current peak picking algorithms (including in Skyline) mis-assign peaks nearly 100% of the time when a higher intensity, isobaric peak exists nearby in time. We highlight that this is because current peak assignment models for each metabolite do not typically have 'knowledge' of elution order for isobaric species; because of the high similarity between metabolites with similar structures, near co-elution is common. We utilize the Hungarian algorithm for multi-assignment problem, along with the relative positions of compounds in the iMT library, to build a peak re-assignment method which removes double-assignments and can accurately select the correct lower intensity isobaric species. This drastically improves automated peak assignment and decreases manual curation time.


[PDF