Predicting GC-MS Biomarker Ratios from FT-ICR MS Molecular Fingerprints by PLS/OPS Regression

Gas chromatography-mass spectrometry (GC-MS) biomarker ratios (e.g., hopanes, steranes, n-alkanes) remain the standard tool for organic geochemical characterization of crude oils, but their determination requires lengthy sample preparation and chromatographic analysis. Fourier transform ion cyclotron resonance mass spectrometry (FT-ICR MS) instead provides an ultrahigh-resolution molecular fingerprint of the whole oil in a single measurement. This study evaluates whether FT-ICR MS molecular compositions can quantitatively predict, by regression, the numerical values of classical GC-MS biomarker ratios, as a potentially faster complement to routine GC-MS quantification. We analyzed 110 crude oils characterized by both electrospray ionization (ESI, negative mode) and atmospheric pressure photoionization (APPI, positive mode) FT-ICR MS (7 T), paired with conventional GC-MS biomarker data. Eighteen diagnostic biomarker ratios were selected as regression targets. For each ratio, partial least squares (PLS) regression models were built using the Ordered Predictors Selection (OPS) algorithm, which iteratively ranks and selects the most informative molecular formulas among the thousands of features detected per ionization mode, reducing model complexity while preserving predictive power. Models were built and validated separately for each heteroatomic-class subset of the ESI and APPI datasets (hydrocarbon and nitrogen-, oxygen- and sulfur-containing compound classes), and for ESI+APPI data-fusion blocks combining both ionization modes. Given the large number of molecular variables relative to sample size, calibration and testing were repeated over 30 independent Monte Carlo cross-validation splits, with variable selection performed anew within each training subset for robust, unbiased performance estimates. Model significance was further confirmed against null models built on randomly permuted response values. Fourteen of the eighteen biomarker ratios yielded statistically validated models, with cross-validated R² ranging from 0.19 to 0.87. Best-performing models included the diasterane/regular sterane ratio (R²=0.87), the 25-norhopane biodegradation fraction (R²=0.81), the Ts/(Ts+Tm) maturity ratio (R²=0.79), the triterpane/sterane ratio (R²=0.75), the sterane/hopane ratio (R²=0.75), the C30-hopane/C27-sterane ratio (R²=0.70), the pristane/phytane ratio (R²=0.65), the C29 sterane isomerization ratio (R²=0.64), and the gammacerane/hopane ratio (R²=0.60). Combining ESI and APPI data improved prediction over either ionization mode alone for most of these ratios. These results demonstrate that a single FT-ICR MS measurement, combined with PLS/OPS regression, can reliably reproduce the numerical values of a wide range of classical hopane-, sterane- and n-alkane-based biomarker ratios, providing a rapid, statistically validated complement to conventional GC-MS quantification in organic geochemistry workflows.