Structure-Preserving Joint Non-negative Tensor Factorization to Identify Reaction Pathways Using Bayesian Networks

Abstract

Extracting meaningful information from spectroscopic data is key to species identification as a first step to monitoring chemical reactions in unknown complex mixtures. Spectroscopic data obtained over multiple process modes (temperature, residence time) from different sensors [Fourier transform infrared (FTIR), proton nuclear magnetic resonance ( 1 H NMR)] comprise hidden complementary information of the underlying chemical system. This work proposes an approach to jointly capture these hidden patterns in a structure-preserving and interpretable manner using coupled non-negative tensor factorization to achieve uniqueness in decomposition. Projections onto the modes of spectral channels, specific to each sensor, are interpreted as pseudo-component spectra, while projections onto the shared process modes are interpreted as the corresponding pseudo-component concentrations across temperature and residence times. Causal structure inference among these pseudo-component spectra (using Bayesian networks) is then used to identify plausible reaction pathways among the identified species representing each pseudo-component. Tensor decomposition of the FTIR data enables the development of reaction sequences based on the identified functional groups, while that of 1 H NMR by itself is lacking in mechanism development as it solely reveals the proton environments in a pseudocomponent. However, jointly parsing spectra from both the sensors is seen to capture complementary information, wherein insights into the proton environment from 1 H NMR disambiguate pseudo-components that have similar FTIR peaks. A scalable method of parallelizing tensor decomposition to handle high-dimensional modes in process data by using grid tensor factorization, while being robust to process data artifacts like outliers, noise, and missing data, has been developed.

Data Fusion-Based Approach for the Investigation of Reaction Networks in Hydrous Pyrolysis of Biomass

“…The kinetics of the decomposition process is not considered explicitly in this work. However, in the context of online reaction monitoring, sophisticated spectroscopic curve resolution algorithms can be used to project real-time spectra onto the temporal data collection mode and the spectroscopic channels, which in accordance with Beer’s law, gain interpretability as the pseudo-component concentrations and pseudo-component spectra, respectively. The kinetics of the underlying chemical transformations can then be assessed from the temporal concentration projections to further facilitate control and optimization .…”

“…A data fusion approach is subsequently used to combine information from the two types of spectroscopic measurements when inferring the chemistry of the process. The essence of data fusion is to link the data from several sensors to carry out deductions that cannot be acquired from a single sensor, as demonstrated by jointly analyzing spectroscopic data to capture complementary information in reactive systems. , Input data from various sources might involve parametric data linked to the object identity, thereby providing a holistic view of the reaction scheme incorporating various distributed sources …”

“…The essence of data fusion is to link the data from several sensors to carry out deductions that cannot be acquired from a single sensor, 14 as demonstrated by jointly analyzing spectroscopic data to capture complementary information in reactive systems. 15,16 Input data from various sources might involve parametric data linked to the object identity, thereby providing a holistic view of the reaction scheme incorporating various distributed sources. 17 A major issue in computational fields like biology or chemical processes is the prevalence of high dimensional datasets to study the network architecture of the variables accurately.…”

Identification of Reaction Network Hypotheses for Complex Feedstocks from Spectroscopic Measurements with Minimal Human Intervention

“…The need for the classifier to be indifferent to the peak height is due to the intensity ambiguity possible in the tensorial decomposition. 17 In the implementation, gas phase FTIR spectra of 11062 molecules were automatically scraped from the NIST database. The International Chemical Identifier for each molecule was used in generating the labels for classification.…”

Prediction of Thermogravimetric Data for Asphaltenes Extracted from Deasphalted Oil Using Machine Learning Techniques

“…In the past 2 decades, RF, SVR, and GBR techniques have become attractive tools for various chemical engineering applications, such as quantitative structure-property relationship development, CO 2 capture, gas chromatography, olefin oligomerization, and development of groundwater potential maps . Though other chemometric methods such as self-modeling curve resolution (SMCR) − and Bayesian learning ,− have been used previously to extract unknown components from bitumen, decision trees and SVR methods have not been applied to TGA data from DAO in the past. RF trees are based on ensemble learning theory and require minimal hypertuning of the parameters as opposed to the choosing and tuning of the weights and number of layers in ANNs.…”