Charged aerosol detector response modeling for fatty acids based on experimental settings and molecular features: a machine learning approach

• DOI: 10.1186/s13321-021-00532-0
• Publication Date: 2021-07-15T11:02:58Z
• Authors: Ruben Pawellek; Jovana Krmar; Adrian Leistner; Nevena Djajić; Biljana Otašević; Ana Protić; Ulrike Holzgrabe
• Keywords: ddc:004; ddc:540; Gradient boosted trees (GBT); Quantitative structure–property relationship modeling (QSPR); High-performance liquid chromatography (HPLC); Information technology; 04 agricultural and veterinary sciences; T58.5-58.64; 541; Chemistry; 0404 agricultural biotechnology; Charged aerosol detector (CAD); Fatty acids; QD1-999; Research Article

Abstract The charged aerosol detector (CAD) is the latest representative of aerosol-based detectors that generate a response independent analytes’ chemical structure. This study was aimed at accurately predicting CAD homologous fatty acids under varying experimental conditions. Fatty from C12 to C18 were used as model substances due semivolatile characterics caused non-uniform behaviour. Considering both conditions and molecular descriptors, mixed quantitative structure–property relationship (QSPR) modeling performed using Gradient Boosted Trees (GBT). ensemble 10 decisions trees (learning rate set 0.55, maximal depth 5, sample 1.0) able explain approximately 99% (Q 2 : 0.987, RMSE: 0.051) observed variance in responses. Validation an external test compound confirmed high predictive ability established (R 0.990, RMSEP: 0.050). With respect intrinsic attribute selection strategy, GBT almost all variables during building. Finally, it attributed highest importance power function value, flow mobile phase, evaporation temperature, content organic solvent phase descriptors such weight (MW), Radial Distribution Function—080/weighted by mass (RDF080m) average coefficient last eigenvector distance/detour matrix (Ve2_D/Dt). identification factors most relevant responsiveness has contributed better understanding underlying mechanisms signal generation. An increased obtained for acetone modifier demonstrated its potential replace more expensive environmentally harmful acetonitrile.