Laser desorption ionization FT-ICR mass spectrometry and CARSPLS for predicting basic nitrogen and aromatics contents in crude oils

• LDI(+)FT-ICR MS and PLS regression with variable selection were used to predict N and aromatics contents.
• The basic nitrogen values ranged from 0.016 to 0.151 wt% and the aromatics values from 8.4 to 35.1 wt%.
• The CARS method was used to select the most informative variables for the PLS calibration model.
• A reduction of 47,873 to 48 and 10 was reached for basic nitrogen and aromatic compounds, respectively.
• Prediction errors were of 0.012% for basic nitrogen and 3.73% for aromatics.

This work describes a methodology for predicting basic nitrogen and aromatics contents in crude oil, using positive ion mode laser desorption ionization coupled to Fourier transform ion cyclotron resonance mass spectrometry (FT-ICR MS) and partial least squares (PLS) regression with variable selection based on competitive adaptive reweighted sampling (CARS), in a procedure called CARSPLS regression. The basic nitrogen and aromatics contents were determined using UOP Method 269-10 and ASTM D5443-14, respectively, for 70 samples of Brazilian crude oil. The basic nitrogen values ranged from 0.016% to 0.151% and the aromatics values from 8.4% to 35.1%. The LDI(+)-FT-ICR mass spectra of crude oil samples have profiles that ranged from m/z 200 to 1000, with an average molar mass distribution (Mw) centered between 554 and 636 Da. Most of the identified organic species are analogues of pyridine and polyaromatic hydrocarbon compounds that belong in the Nx, Nx[H], HC, and HC[H] classes. Additionally, the number of detected Nx species increases systematically with increasing basic nitrogen and aromatics content. The CARS method was used to select the most informative variables for the PLS calibration model, thereby reducing the number of variables from 47,873 to 48 and 10 for basic nitrogen and aromatic compounds, respectively. The prediction model based on CARSPLS presented RMSEP of 0.012 and 3.73 wt% for basic nitrogen and aromatics contents, respectively. In addition, it was possible to verify that the accuracy of CARSPLS is ever better than only using PLS regression (with all variables). The application of CARSPLS algorithm is therefore justified by easier interpretation of the FT-ICR MS data after variables reduction.