Correction of mass spectral drift using artificial neural networks

Area of Science:

• Analytical Chemistry
• Biotechnology
• Spectroscopy

Background:

• Routine identification of microorganisms and quantification of determinands using pyrolysis mass spectrometry (PyMS) relies on comparing new spectra to existing libraries.
• Mass spectral fingerprints often lack long-term reproducibility due to instrumental drift, hindering the use of previously developed models for new analyses.

Purpose of the Study:

• To address the challenge of instrumental drift in PyMS and other analytical techniques.
• To enable the reliable use of multivariate calibration models for analyzing new data despite instrumental variations over time.

Main Methods:

• Utilizing neural networks to model and correct for instrumental drift in pyrolysis mass spectrometry.
• Applying multivariate calibration models to relate mass spectra to biological features of interest.

Main Results:

• Identified that instrumental drift significantly impacts the long-term reproducibility of PyMS data.
• Demonstrated that neural networks can effectively correct for PyMS instrumental drift.
• Showed that corrected data allows previously developed models to accurately estimate determinand concentrations and bacterial identities from new spectra.

Conclusions:

• Neural networks offer a robust solution for correcting instrumental drift in PyMS.
• This drift correction method enhances the reliability of PyMS for routine identification and quantification.
• The approach is broadly applicable to various analytical techniques susceptible to instrumental drift.