東華大學圖書館 |

Bridging the Gap Between Chemical Analysis and Applied Statistics to Expand the Applications of Atomic Spectrometric Techniques.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Bridging the Gap Between Chemical Analysis and Applied Statistics to Expand the Applications of Atomic Spectrometric Techniques./
作者:	Carter, Jake Alexander.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2020,
面頁冊數:	296 p.
附註:	Source: Dissertations Abstracts International, Volume: 82-03, Section: B.
Contained By:	Dissertations Abstracts International82-03B.
標題:	Analytical chemistry. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28087519
ISBN:	9798672146157

Bridging the Gap Between Chemical Analysis and Applied Statistics to Expand the Applications of Atomic Spectrometric Techniques.
Carter, Jake Alexander.

Bridging the Gap Between Chemical Analysis and Applied Statistics to Expand the Applications of Atomic Spectrometric Techniques. - Ann Arbor : ProQuest Dissertations & Theses, 2020 - 296 p.

Source: Dissertations Abstracts International, Volume: 82-03, Section: B.

Thesis (Ph.D.)--Wake Forest University, 2020.

This item must not be sold to any third party vendors.

The combination of (trace) elemental analysis with data science techniques is an important contribution to atomic spectrometry and its application to a broad range of research fields. With temperatures as high as 10,000 K, the inductively coupled plasma (ICP) is a robust source of atomization, excitation and ionization, capable of providing trace level information for the vast majority of elements on the periodic table. Recently, data science has catalyzed research across science by helping to process and interpret large amounts of complex data acquired from instrumental techniques. The proposed body of work shows the merit of using data-driven algorithms to enhance the analytical capabilities of modern atomic spectrometry in two ways: (i) furthering the interpretation and use of data acquired using traditional calibration methods for environmental and health-related applications, and (ii) identifying and resolving matrix effects that can compromise analyses based on the traditional external standard calibration method (EC).Undoubtedly, ICP-based techniques are the most sensitive strategies for trace level analysis of metals and non-metals (e.g. Cu, Pb, S, and Zn). Considering the importance of environmental and biological health, there is a need for data analysis frameworks connecting raw concentrations from ICP analysis to interpretable data exploration results for health-related experts. For this purpose, we have used supervised (e.g. random forest) and unsupervised (e.g. principal component analysis) learning strategies, applied to atomic spectrometry data, to help ensure the safety of school drinking water and to identify type-2 diabetes.Using similar statistical techniques, data analysis workflows have been developed to evaluate, in real time, the applicability of EC for a given sample matrix. The severity of matrix effects were described in principal component space by the distance between simple-matrix solutions and complex-matrix samples. Matrix severity was also visualized using other modern unsupervised leaning methods such as t-distributed stochastic neighbor embedding (t-SNE), and uniform manifold approximation and projection (UMAP). In addition, an estimation of analytical accuracy was determined from tuned supervised learning models (e.g. pls and glmnet). Further, signal bias correction featuring naturally occurring, plasma-based species were proposed for sample matrices not suitable for EC.Each of the illustrated works first describe an optimized chemical analysis process prior to the consideration of advanced data analysis strategies. The amount of information extracted from advanced data processing is fundamentally limited to the quality of the data available, where the quality of data is dependent on the optimization of sampling, sample preparation and instrumental analysis. Characteristics of certain atomic spectrometry methods may hinder the use of data science techniques if not accounted for. To properly train supervised learning models, especially for small datasets (commonly found in atomic spectrometry as a result of the sample preparation bottleneck), it is imperative that proper resampling and data partition strategies are used to evaluate potential overfitting and to get an estimate of predictive accuracy for a given model. Similarly, the merit of using more complex models should be weighed against the use of simpler models that are easier to interpret.

ISBN: 9798672146157Subjects--Topical Terms:

3168300
Analytical chemistry.
Subjects--Index Terms:

Drinking water

Bridging the Gap Between Chemical Analysis and Applied Statistics to Expand the Applications of Atomic Spectrometric Techniques.
LDR:04712nmm a2200385 4500 001 2344757
005 20220531062131.5
008 241004s2020 ||||||||||||||||| ||eng d
020 $a 9798672146157
035 $a (MiAaPQ)AAI28087519
035 $a AAI28087519
040 $a MiAaPQ $c MiAaPQ
100 1 $a Carter, Jake Alexander. $3 3683562
245 1 0 $a Bridging the Gap Between Chemical Analysis and Applied Statistics to Expand the Applications of Atomic Spectrometric Techniques.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2020
300 $a 296 p.
500 $a Source: Dissertations Abstracts International, Volume: 82-03, Section: B.
500 $a Advisor: Jones, Bradley T.;Donati, George L.
502 $a Thesis (Ph.D.)--Wake Forest University, 2020.
506 $a This item must not be sold to any third party vendors.
520 $a The combination of (trace) elemental analysis with data science techniques is an important contribution to atomic spectrometry and its application to a broad range of research fields. With temperatures as high as 10,000 K, the inductively coupled plasma (ICP) is a robust source of atomization, excitation and ionization, capable of providing trace level information for the vast majority of elements on the periodic table. Recently, data science has catalyzed research across science by helping to process and interpret large amounts of complex data acquired from instrumental techniques. The proposed body of work shows the merit of using data-driven algorithms to enhance the analytical capabilities of modern atomic spectrometry in two ways: (i) furthering the interpretation and use of data acquired using traditional calibration methods for environmental and health-related applications, and (ii) identifying and resolving matrix effects that can compromise analyses based on the traditional external standard calibration method (EC).Undoubtedly, ICP-based techniques are the most sensitive strategies for trace level analysis of metals and non-metals (e.g. Cu, Pb, S, and Zn). Considering the importance of environmental and biological health, there is a need for data analysis frameworks connecting raw concentrations from ICP analysis to interpretable data exploration results for health-related experts. For this purpose, we have used supervised (e.g. random forest) and unsupervised (e.g. principal component analysis) learning strategies, applied to atomic spectrometry data, to help ensure the safety of school drinking water and to identify type-2 diabetes.Using similar statistical techniques, data analysis workflows have been developed to evaluate, in real time, the applicability of EC for a given sample matrix. The severity of matrix effects were described in principal component space by the distance between simple-matrix solutions and complex-matrix samples. Matrix severity was also visualized using other modern unsupervised leaning methods such as t-distributed stochastic neighbor embedding (t-SNE), and uniform manifold approximation and projection (UMAP). In addition, an estimation of analytical accuracy was determined from tuned supervised learning models (e.g. pls and glmnet). Further, signal bias correction featuring naturally occurring, plasma-based species were proposed for sample matrices not suitable for EC.Each of the illustrated works first describe an optimized chemical analysis process prior to the consideration of advanced data analysis strategies. The amount of information extracted from advanced data processing is fundamentally limited to the quality of the data available, where the quality of data is dependent on the optimization of sampling, sample preparation and instrumental analysis. Characteristics of certain atomic spectrometry methods may hinder the use of data science techniques if not accounted for. To properly train supervised learning models, especially for small datasets (commonly found in atomic spectrometry as a result of the sample preparation bottleneck), it is imperative that proper resampling and data partition strategies are used to evaluate potential overfitting and to get an estimate of predictive accuracy for a given model. Similarly, the merit of using more complex models should be weighed against the use of simpler models that are easier to interpret.
590 $a School code: 0248.
650 4 $a Analytical chemistry. $3 3168300
650 4 $a Artificial intelligence. $3 516317
650 4 $a Water resources management. $3 794747
653 $a Drinking water
653 $a Inductively coupled plasma
653 $a Lead
653 $a Machine learning
653 $a Matrix effects
653 $a Non-invasive diagnostics
690 $a 0486
690 $a 0800
690 $a 0595
710 2 $a Wake Forest University. $b Chemistry. $3 2102793
773 0 $t Dissertations Abstracts International $g 82-03B.
790 $a 0248
791 $a Ph.D.
792 $a 2020
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28087519