Quantitative analysis of ToF-SIMS data of a two organic compound mixture using an autoencoder and simple artificial neural networks

Rationale: Matrix effects cause a nonlinear relationship between ion intensities and concentrations in mass spectrometry, including time-of-flight secondary ion mass spectrometry (ToF-SIMS). Here, two artificial neural network (ANN)-based methods, autoencoder-based and simple ANN methods, were employed for the quantitative and qualitative analyses of a two organic compound mixture via ToF-SIMS. Methods: The multilayer model sample contained a mixture of Irganox 1010 and Fmoc-pentafluoro-L-phenylalanine (Fmoc-PFLPA). The sample's positive and negative ion depth profiles were collected through ToF-SIMS. ToF-SIMS-derived cross-sectional image datasets were analyzed using three unsupervised methods, namely principal component analysis (PCA), multivariate curve resolution (MCR), and use of a sparse autoencoder (SAE). The supervised simple ANN method was optimized based on the spectra and validated by predicting the test dataset ratios of Irganox 1010. Results: The results obtained using the SAE demonstrated linear calibration curves and appropriate material distribution images. The Irganox 1010 and Fmoc-PFLPA positive and negative ion datasets exhibited >0.97 correlation coefficients. The PCA and MCR results demonstrated lower linearity than that of SAE. Moreover, SAE weights indicated the ions important for each organic compound. The simple ANN method accurately predicted the ratios in the test dataset and indicated the important ions. Conclusions: Both the supervised and unsupervised methods based on ANN, which were employed in regulating nonlinear relationships, were effective in the quantitative and qualitative analyses of the ToF-SIMS data of the two organic compound mixtures. Regarding qualitative analysis, both ANN-based methods indicated specific ions from the molecules in the sample.