Assessment of artificial neural network to identify compositional differences in ultrahigh-resolution mass spectra acquired from coal mine affected soils

This study assessed the applicability of artificial neural networks (ANNs) as a tool to identify compounds contributing to compositional differences in coal-contaminated soils. An artificial neural network model was constructed from laser desorption ionization ultrahigh-resolution mass spectra obtained from coal contaminated soils. A good correlation (R2 = 1.00 for model and R2 = 0.99 for test) was observed between the measured and predicted values, thus validating the constructed model. To identify chemicals contributing to the coal contents of the soils, the weight values of the constructed model were evaluated. Condensed hydrocarbon and low oxygen containing compounds were found to have larger weight values and hence they were the main contributors to the coal contents of soils. In contrast, compounds identified as lignin did not contribute to the coal contents of soils. These findings were consistent with the conventional knowledge on coal and results from the conventional partial least square method. Therefore, we concluded that the weight interpretation following ANN analysis presented herein can be used to identify compounds that contribute to the compositional differences of natural organic matter (NOM) samples. [Display omitted] •Molecular differences in coal-contaminated soils studied with ANN and FT-ICR MS.•First study to use ANN analysis to study NOM spectra obtained from FT-ICR MS.•Weight analysis applied identify compounds in FT-ICR MS spectra.•Compounds identified as lignin did not contribute to the coal contents of soil.•Condensed hydrocarbon and low oxygenated compounds contributed to coal in soil.