Loading...

Evaluating the Ability of Deep Learning Algorithms to Improve Qualitative and Quantitative Chromatographic Analyses

Piltan Eraghi, Mahsa | 2023

101 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55888 (03)
  4. University: Sharif University of Technology
  5. Department: Chemistry
  6. Advisor(s): Parastar Shahri, Hadi
  7. Abstract:
  8. In recent years, comprehensive two-dimensional gas chromatography (GC×GC) has been widely used, especially in the separation and measurement of complex mixtures, due to its features such as higher peak capacity and better resolution in separation than usual chromatography methods, which the mixture analysis of polycyclic aromatic hydroarbons (PAHs) compounds have been targeted in this study. One of the main challenges in this field is the large amount of data and the qualitative and quantitative interpretation of the resulting data. Therefore, the identification, separation and quantitative measurement of species in this field is important, especially in the presence of interference in the sample matrix. In this research, with the aim of quantification in comprehensive two-dimensional gas chromatography, two strategies have been proposed. In the first strategy using the concept of two-dimensional chromatography, the use of total ion chromatogram (TIC) was proposed for the quantitative analysis of target analytes including dibenzoanthracene, fluorene, dibenzothiophene, naphthalene, 1-methylnaphthalene, 3,6-dimethylphenanthrene, pyrene, 1-methylphenanthrene, phenanthrene and anthracene. In this way, the data related to different cuts of the target species that are injected from the first column to the second chromatography column, are placed in a new matrix and the concentration of each cut was determined based on the ratio of the total concentration of that species in the first column. Then, using conventional multivariate calibration methods such as partial least square regression (PLSR), support vector machine (SVM) with different kernels and radial basis function-artificial neural network (RBF-ANN) method, the relevant model was built and analytical figures of merit (AFOMs) were obtained for them. For example, the values of sensitivity and regression coefficient obtained for analytes in the PLSR method are in the range of 2.64×105-2.41×106 and 0.991-0.999, respectively.Also, this idea can be extended to other chromatographic methods such as liquid chromatography and compared to the conventional method based on the use of the entire chromatogram in the form of pixels with data volume reduction methods, it has the advantage that this idea is based on the concept of two-dimensional chromatography itself and it does not need any pre-processing methods to reduce data volume. In the second strategy with the aim of considering the alignment of the mass spectrum in quantitative analysis of target analytes including naphthalene, 1-methylnaphthalene, acenaphthene, acenaphthylene, fluorene, phenanthrene, pyrene, dibenzothiophene, dimethylnaphthalene, benzofluoranthene, chrysene, benzonaphthothiophene and fluoranthene and also improving the conventional process of analyzing such data, which is the analysis of part by part of the chromatogram, the whole chromatogram was analyzed by the multivariate curve resolution-alternating least squares (MCR-ALS) method, and then univariate and multivariate figures of merit were obtained for the target analytes. For example, the sensitivity values in multivariate calibration for analytes are in the range of 5.40×108-2.70×1010. Since the volume of data for all species in the calibration concentration range is very large, the wavelet transform (WT) method was used in the chromatographic alignment to reduce the volume of data. Also, singular value decomposition (SVD) was used to determine the number of species and orthogonal projection approach (OPA) method was used for the initial guess of spectral profiles. Non-negativity and normalization (spectral alignment) constraints were also used during ALS optimization. In order to check the effectiveness of the two strategies mentioned in real samples, the identification and quantitative measurement of PAHs in the aromatic part of the heavy oil sample analyzed by GC×GC-TOFMS was used and the results show the appropriate performance of the proposed methods. Also, the two proposed strategies are used in different fields based on the desired goals
  9. Keywords:
  10. Comprehensive Two Dimensional Gas Chromatography ; Machine Learning ; Multivariate Curve Resolution ; Polynuclear Aromatic Hydrocarbons (PAHs) ; Chemometrics Method ; Deep Learning

 Digital Object List

 Bookmark

No TOC