Meet the Researcher - Robert Spiers

Breakout Room: 6

RobertSpiers_HeadshotResearcher Name: Robert Spiers
Title of Research: Advancing Chemical Analysis for the Consumer and Industry
Division Representing: Chemistry
Institution: Idaho State University
Institution Location: Idaho
Home State: Idaho
District Number: 2
Advisor/Mentor: John Kalivas
Funding Source: National Science Foundation

Research Experience:  
Robert Spiers is in his third year of studying Physics at Idaho State University. Also included in his studies are minors in Computer Science and Mathematics, for a five-semester cumulative GPA of 3.96. As a member of the university honors program, he was able to take honors chemistry with Dr. John Kalivas, who offered Spiers a position to perform undergraduate research in Analytical Chemistry. Starting in April 2019, Spiers learned the fundamentals of chemometrics—the field of chemistry which integrates mathematics and statistical methods with chemical data analysis, now commonly referred to as machine learning. Research started in June of 2019 culminating in two manuscripts, one of which is under review by the Journal of Chemical Information and Modeling. The other manuscript is in its final stages of revision for submission to Analytical Chemistry and he is first author on both manuscripts. Spiers has made five posters and two oral meeting presentations spanning his research. He began working on a second project in March 2020 which focuses on calibration sample selection to improve model predictions. In the future, Spiers intends to pursue graduate school for Physics, where he will study the applications of computational methods in condensed matter and plasma physics. His undergraduate research experience will be invaluable, as it contributed not only in his ability to perform research at a high level, but also to apply his knowledge of machine learning, statistics, chemistry, and spectroscopy into his physics education.

Presentation Experience: 
Having been working as an undergraduate researcher for almost two years, Spiers has attended six academic conferences. He presented posters at the Idaho Conference for Undergraduate Research (ICUR) for both 2019 and 2020, where the audience had little to no knowledge of chemistry. Spiers has also presented posters at the FACSS SciX conference for 2019 and 2020, a highly technical conference for chemical spectroscopy that contains chemometric sessions. Spiers won first prize for the student poster award at that conference in 2019, where most posters were presented by eligible graduate students. He also presented a poster at the Idaho State University Undergraduate Research Symposium in 2020. Spiers was offered the honor to present a full length talk at the ACS National Meeting in August of 2020, where he delivered a 25-minute-long talk during the broadcasted session of the meeting. A few months later, he gave another talk in the invited chemometric symposium titled Advances in Multivariate Calibration” at the FACSS SciX meeting. Between presenting five posters and giving two talks, Spiers has extensive experience with relaying scientific information to both academic and non-academic audiences.

Significance of Research:       
Technological advancements have yielded procedural accelerations in many fields. However, direct substance composition analysis by classical laboratory methods (e.g. determining protein content in meats) is an element of chemistry physically limited in speed. Instead, machine learning methods harness spectroscopy to rapidly determine composition. However, accuracy of these methods is related to the degree of difference between original (primary) calibration sample and spectral measurement conditions and new (secondary) prediction conditions. Model updating facilitates this transfer between pristine laboratory primary conditions and field secondary conditions, such as changes in instrument type or growing season. Updating methods commonly require time-consuming composition analysis for a few secondary samples (labeled samples), however our mechanism does not. The presented framework is two-fold: a large range of models are generated by nulling each model to the differences between measurement conditions, then a subset of accurate models is identified by using the novel model diversity and prediction similarity (MDPS) criterion we developed. Results show that the models developed and selected using this framework rival the accuracy of expensive modeling methods: those requiring a few labeled secondary samples and those using many labeled secondary samples to calibrate an entirely new model. Of the current updating techniques employed in chemical modeling, this project provides the first mechanism which is accurate, robust, procedurally complete, and does not require labeled secondary samples. Thus, this method is ideal for public usage; any person can use a handheld spectrometer, upload data to cloud computing, and within seconds receive the protein content of their food.

Uniqueness of Research: 
Presented is a hands-off chemical modeling approach allowing for rapid, inexpensive, accurate chemical analyses via spectroscopy. This robust method requires no information about prediction sample and measurement conditions. Algorithm applications include handheld spectral analysis of pharmaceuticals, smart farming, food adulteration and authentication as well as in-line industrial process analyzers. Forensic applications are abundant, such as crime scene field analysis.