Year: 2025
Type: article
Abstract: Chemical isotope patterns provide crucial molecular fingerprints in mass spectrometry, yet their discrete nature limits integration with modern machine learning frameworks. While research has shown the importance of isotope patterns for compound identification, little attention has been paid to developing continuous representations suitable for computational analysis. This article examines if a relationship exists between Gaussian modeling parameters and information preservation in isotope pattern transformation, and if any such relationship can be attributed to molecular characteristics or methodological factors. A combination of theoretical calculations and statistical validation were used from 1,902 compounds acquired from PubChem. Findings showed significant relationships between Gaussian modeling parameters (σ = 0.02 Da, 1000-dimensional vectors) and successful pattern conversion with >99.5% information preservation. Multivariate analyses indicated that these relationships could be explained by differences in molecular composition in terms of size and structural complexity variables. Principal component analysis revealed that 85.2% of variance could be explained by the first 10 components, with PC1 correlating with molecular weight (r=0.89) and PC3 with structural complexity (r=0.68). These results demonstrate the importance of considering molecular diversity in analyses of isotope pattern modeling, and controlling for chemical structure effects in pattern transformation.
Author: Sravya Varanasi
Institution: Cornell University
Language: English
Subfield: Spectroscopy
Field: Chemistry
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4411277487