Molecular-Level Prediction of Asphaltene Precipitation by Integrating APPI (+) FT-ICR Mass Spectrometry and Machine Learning

Asphaltenes are a complex and polydisperse fraction of crude oil, primarily characterized by their insolubility in n-alkanes and solubility in toluene. Their precipitation during production and processing can cause severe flow assurance problems, leading to increased operational costs and reduced refining efficiency. Although experimental techniques are widely employed to evaluate asphaltene stability, predicting precipitation behavior from the molecular composition of crude oils remains a significant challenge due to their complex chemical composition. In this context, ultra-high-resolution mass spectrometry (UHRMS), particularly Fourier transform ion cyclotron resonance mass spectrometry (FT-ICR MS), combined with machine learning has been used as a promising approach for correlating molecular fingerprints with asphaltene precipitation behavior. This study investigated the molecular classes obtained from Atmospheric Pressure Photoionization on positive mode, APPI (+) FT-ICR MS data, and identified those most strongly associated with asphaltene precipitation behavior in crude oil samples. A total of 19 crude oil samples provided by Petrobras were classified according to the results of the isothermal depressurization test for asphaltene precipitation: 12 non-precipitated (NP) and 7 precipitated (P). Each sample was prepared at a concentration of 500 mg L⁻¹ in toluene:methanol (1:1, v/v) and analyzed by APPI (+) FT-ICR MS using a 7T SolariX 2xR (Bruker Daltonics, Bremen, Germany), utilizing a krypton UV lamp with an energy of 10.6 eV. Spectra were acquired with 300 scans over a mass range of 150-1200. Molecular formula assignments and data processing were performed by DataAnalysis software (Bruker, Billerica, MA, USA), Composer64 v1.5.3 software (Sierra Analytics Inc., Modesto, CA, USA), and MATLAB 2025a (MathWorks, Natick, MA, USA). A supervised Partial Least Squares Discriminant Analysis (PLS-DA) model was trained using the 19 crude oil samples, with the molecular classes previously identified in a proof-of-concept study as relevant predictors of asphaltene stability (HC, N, NO, O, OS, and S classes) as input variables. The PLS-DA model showed high discriminative ability, classifying correctly 16 of the 19 crude oil samples with known precipitation behavior, corresponding to an accuracy of approximately 85%. The model showed 90.9% sensitivity for NP samples and 75% specificity for P samples. These results highlight the potential of APPI (+) FT-ICR MS molecular fingerprints for the preliminary assessment of asphaltene precipitation behavior without relying on experimental stability measurements.