De Novo PFAS Annotation and Classification Using Highly Accurate Formula Prediction and Kaufmann Algorithms Embedded in FluoroMatchSuite

Importance of the Topic

Per- and polyfluoroalkyl substances (PFAS) are pervasive environmental contaminants with growing regulatory and health implications. Accurate, comprehensive profiling of PFAS mixtures is essential for risk assessment, remediation efforts, and regulatory compliance. Traditional non-targeted workflows relying on MS/MS fragmentation data can suffer from incomplete coverage and ambiguous identifications, especially in complex environmental matrices. Developing robust MS1-only tools enhances detection sensitivity and formula assignment accuracy across diverse sample types.

Study Objectives and Overview

This study introduces two novel MS1-based algorithms embedded in the FluoroMatch Suite to complement existing MS/MS tools: formula prediction via high-precision isotope and mass defect filtering, followed by homologous series voting; and PFAS classification using Kaufmann kernel density analysis. Validation was performed on two NIST interlaboratory samples: NIST A (12 known PFAS standards) and NIST C (AFFF-contaminated soil extract), benchmarking false positive and negative rates against reference assignments.

Methodology and Instrumentation

The FluoroMatch workflow begins with full-scan LC-HRMS data acquisition in either DDA or DIA mode. Data are imported from vendor formats or mzML into MassHunter Explorer or MZMine 4 for peak picking and blank filtration.

Instrumentation:

• Liquid chromatography coupled to high-resolution mass spectrometry (LC-HRMS/MS)
• Data-dependent acquisition (DDA) and data-independent acquisition (DIA) methods
• Software tools: MassHunter Explorer, MZMine 4, R packages, and FluoroMatch Suite

Main Results and Discussion

Formula Prediction Algorithm:
The algorithm isolates the highest-intensity MS1 spectra, applies mass defect filtering to remove non-PFAS signals, and enforces isotopic pattern quality checks. Candidate formulas pass atomic constraints (senior rule, nitrogen rule, element ratios) before scoring by m/z and intensity similarity. Homologous series voting consolidates assignments across repeating CF₂ units, reducing ambiguity.

• NIST A: initial false positive rate 17%, reduced to 0% with series voting; false negative rate rose from 0% to 17% without voting but remained 0% with voting.
• NIST C: initial false positive rate 26%, reduced to 0% with voting; false negative rate improved from 30% to 6% with voting.

Kaufmann Classification Algorithm:
Kernel density of EPA PFAS reference data in a mass defect versus m/z space defines isolines at 95% coverage. Unknown features are mapped to this space, with thresholds determining PFAS likelihood.

• NIST A: 24% of non-PFAS features misclassified as PFAS; zero confirmed PFAS were missed.
• NIST C: only 4% of non-PFAS misclassified; 6% of true PFAS fell outside the 95% isoline.
• Approximately 15% of features lacked an M+1 isotopic peak and were unclassified.

Benefits and Practical Applications

Integrating MS1-only modules into non-targeted PFAS workflows enhances formula coverage and PFAS classification specificity without additional MS/MS experiments. Laboratories can achieve near-zero false positives in formula assignments and maintain high PFAS detection confidence in complex matrices such as firefighting foam–contaminated soils. The streamlined approach reduces analysis time and supports regulatory monitoring initiatives.

Future Trends and Opportunities

Advancements may include machine learning models trained on expanded PFAS libraries to refine classification boundaries and improve predictions for novel chemistries. Integration with real-time processing platforms could enable on-site environmental screening. Expanding homologous series voting to multi-dimensional chromatographic separation will further resolve isomeric PFAS compounds.

Conclusion

The FluoroMatch Suite’s MS1-only formula prediction and Kaufmann classification algorithms deliver highly accurate, low-false-positive PFAS annotations. Homologous series voting is critical for eliminating misassignments, and Kaufmann isoline filtering effectively discriminates PFAS from background interferences. Combined with existing MS/MS tools, these innovations strengthen non-targeted PFAS workflows for research, environmental monitoring, and regulatory compliance.

References

Agilent Technologies, Inc. “De Novo PFAS Annotation and Classification Using Highly Accurate Formula Prediction and Kaufmann Algorithms Embedded in FluoroMatch Suite.” ASMS 2025 Poster WP 113, May 15, 2025.