Metabolomics: New peak detection (Pyco) and isotope grouping (Prism) algorithms for an improved compound detection workflow

Significance of the topic

The reliable detection and grouping of chromatographic peaks and their isotopes is fundamental for accurate compound identification and quantification in untargeted LC/MS analyses

Objectives and study overview

This work introduces two novel, parameter-free algorithms implemented in Thermo Scientific™ Compound Discoverer™ 3.3: Pyco for peak detection and Prism for isotope grouping
It aims to overcome limitations of model-based approaches, increase sensitivity, reduce false positives, and shorten computational time across various chromatography methods

Methodology

Four-step detection workflow:

1. Trace generation and filtering with interpolation of small gaps (max two missing points)
2. Peak detection using Pyco, which locates local inflection points and selects significant features without assuming Gaussian shape
3. Isotope pattern identification via Prism, matching mass shifts for C, H, O, N, S, Br, Cl
4. Adduct and fragment grouping based on coelution and mass relationships

Key signal processing steps include baseline correction using AirPLS (adaptive iteratively reweighted penalized least squares), total variation denoising to preserve peak shape, and Gaussian smoothing

Instrumentation used

Thermo Scientific™ Compound Discoverer™ 3.3 software performed processing on raw LC/MS data
Hardware reference: HP Z840 workstation, dual Intel Xeon E5-2667 CPUs (3.2 GHz, 8 cores each), 64 GB RAM

Main results and discussion

Comparison of Compound Discoverer versions:

• Version 3.3 with Pyco/Prism detected 104 332 compounds (peak rating ≥ 4), reducing false positives to 19 931, run time 4 h 27 min
• Legacy version 3.2 found 59 958 compounds, no filtering by quality, run time 17 h 3 min

Quantitative reproducibility evaluated on 1100 spiked standards showed significantly lower CVs for Pyco compared to legacy and comparable to top external tools
Peak quality factors (jaggedness, zig-zag index, modality, FWHM-to-base ratio, CV, relative area) deliver a composite Peak Rating for filtering

Benefits and practical applications

• Improved sensitivity: more true compounds and isotopic features detected
• Reduced false positives via Peak Rating filtering
• Faster processing enables larger studies and high-throughput projects
• Model-free approach adapts to diverse peak shapes and chromatographic conditions

Future trends and potential applications

Anticipated developments include integrating machine learning classifiers to refine peak quality assessment, expanding support for isotopic labeling studies, deploying real-time peak detection in process monitoring, and cloud-based scalable workflows for large cohort studies

Conclusion

The Pyco and Prism algorithms deliver a robust, parameter-free solution for peak and isotope detection, enhancing both analytical performance and computational efficiency in untargeted metabolomics and small-molecule studies

References

• Z.-M. Zhang, S. Chen, Y.-Z. Liang. Baseline correction using adaptive iteratively reweighted penalized least squares. Analyst 135(5):1138–1146 (2010).
• Pandey et al. MetaClean: a machine learning-based classifier for reduced false positive peak detection in untargeted LC–MS metabolomics data. Metabolomics (2020).
• Li et al. Comprehensive evaluation of untargeted metabolomics data processing software in feature detection, quantification and discriminating marker selection. Anal. Chimica Acta (2018).