Modeling the adsorption of PAH mixture in silica nanopores by molecular dynamic simulation combined with machine learning.

The persistence of polycyclic aromatic hydrocarbons (PAHs) in contaminated soils is largely controlled by their molecular fate in soil pores. The adsorption and diffusion of 16 PAHs mixture in silica nanopore with diameter of 2.0, 2.5, 3.0 and 3.5 nm, respectively, were characterized by adsorption energy, mean square displacement, free surface area and free volume fraction using molecular dynamic (MD) simulation. Results suggested that PAHs adsorption in silica nanopores was associated with diffusion process while competitive sorption was not the dominant mechanism in context of this study. The partial least squares (PLS) regression and machine learning (ML) methods (i.e. support vector regression, M5 decision tree and multilayer perceptrons) were used to correlate the adsorption energy with the pore diameter and PAH properties (number of carbon atoms, aromatic ring number, boiling point, molecular weight, octanol-water partition coefficient, octanol-organic carbon partition coefficient, solvent accessible area, solvent accessible volume and polarization). Results indicated that the PAH adsorption could not be predicted by linear regression as the R(2)Y and Q(2)Y coefficients of PLS analysis was 0.375 and 0.199, respectively. The nonlinearity was well recognized by ML with correlation coefficient up to 0.9. Overall, the combination of MD simulation and ML approaches can assist in interpreting the sequestration of organic contaminants in the soil nanopores.