Professor Facundo M. Fernández

Facundo M. Fernández
Regents’ Professor and Vasser-Woolley Chair in Bioanalytical Chemistry, School of Chemistry and Biochemistry, Georgia Institute of Technology
Abstract

Machine Learning in 21st Century Analytical Chemistry

Alexandria Van Grouw ¹, Carter Asef ¹, Olatomiwa Bifarin ¹, Priyanka Priyadarshani ², Samuel Moore ³, David Gaul ^1,3, Molly Ogle ⁴, Luke Morgensen ², Johnna Temenoff ⁴, Carlos A. Saavedra-Matiz ⁵, Joseph J. Orsini ⁵, Konstantinos Petritis ⁶, Melissa Kemp ⁴, Facundo M. Fernandez ^1,3,*

¹ School of Chemistry and Biochemistry, Georgia Institute of Technology.
² School of Chemical, Materials, and Biomedical Engineering, University of Georgia.
³ Petit Institute of Bioengineering and Bioscience, Georgia Institute of Technology.
⁴ The Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of Technology & Emory University.
⁵ Newborn Screening Program, Wadsworth Center, New York State Department of Health.
⁶ Division of Laboratory Sciences, National Center for Environmental Health, Centers for Disease Control and Prevention.
* [email protected]

Advances in analytical chemistry instrumentation and computational tools have now made it possible to examine biological processes in complex systems with an exquisite level of detail. This includes collections of analytes (or “omes”) with a given specific nature such as the transcriptome (nucleic acids), the proteome (proteins and peptides) and the metabolome (metabolites and lipids). Of these, the metabolome is the most sensitive to perturbations and interventions, yielding a molecular profile that is closest to the physiological phenotype. Metabolomic profiles are therefore sensitive to reprogramming observed in early disease stages and disease progression, which are more difficult to detect at the proteome or transcriptome levels. 

While mapping the metabolome with high resolution techniques such as liquid chromatography, nuclear magnetic resonance spectroscopy and mass spectrometry, generation of highly dimensional and complex datasets is the norm, rather than the exception. Machine learning (ML) is therefore used in metabolomics to analyze and interpret these vast amounts of complex data, with tasks including classification, regression, clustering, dimensionality reduction and anomaly detection, among others. Despite its power, ML is not without limitations. These include Data Dependency (ML models heavily rely on the quality and quantity of data available for training. Biased or incomplete data can lead to inaccurate or unfair predictions); Interpretability (“ome” machine learning models, like deep neural networks, can be complex and challenging to interpret, this lack of transparency can be a drawback in fields where explainability is crucial, such as healthcare), and Overfitting (models that are overly complex or trained on limited data may memorize the training examples instead of learning generalizable patterns, this can lead to poor performance on unseen data). 

In this presentation, I will discuss the analytical and ML tools we have developed for single cell mass spectrometry, present a ML-enabled metabolomics workflow to improve newborn screening, and discuss how ML can be used to improve quantitation in lipidomics when chemical standards are not available. I will also showcase recent results on how explainable artificial intelligence (XAI) is helping solve the interpretability issue in the context of clinical metabolomics, and how natural language processing (NLP) can guide the interpretation of the vast amounts of scientific knowledge being generated in our field in a way that is both meaningful and useful.

Facundo M. Fernández

Prof. Facundo M. Fernández is the Regents’ Professor and Vasser-Woolley Chair in Bioanalytical Chemistry in the School of Chemistry and Biochemistry at the Georgia Institute of Technology. He received his BSc and MSc in Chemistry from the Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires in 1995, and his PhD in Analytical Chemistry from the same University, in 1999. Between 2000 and 2001, he was a postdoc in the research group of Richard N. Zare in the Department of Chemistry at Stanford University. Between 2002- 2003, he joined the group of Vicki Wysocki in the Department of Chemistry at the University of Arizona as a senior postdoc and then research scientist. Prof. Fernandez is internationally renowned for his work in bioanalytical chemistry, with his research focusing on the development of new tools for assaying small volume samples, tissues, and single cells, and applying such methods to better understanding diseases such as cancer, CF and IBD. He is the author of 220+ peer- reviewed publications, has presented 225+ invited lectures, and graduated 31 Ph.D. and M.Sc. students. He is also the academic director for the Systems Mass Spectrometry Core (SyMS-C) at the Parker H. Petit Institute for Bioengineering and Bioscience at Georgia Tech, where he oversees a portfolio numerous mass spectrometers from most major vendors. He has received several awards, including the NSF CAREER award, the CETL/BP Teaching award, the Ron A. Hites best paper award from the American Society for Mass Spectrometry, and the Beynon award from Rapid Communications in Mass Spectrometry, among others. He serves on the editorial board of The Analyst and as an Associate editor for the Journal of the American Society for Mass Spectrometry and Frontiers in Chemistry. His current research team of 15-20 people is interested in metabolomics, development of new ionization sources, MS imaging, machine learning and ion mobility spectrometry. The research is supported by agencies such as NIH, NSF, NASA, IARPA and DoD. In his free time, he enjoys camping and off-roading with his family, kayaking, and climbing summits to connect with other nerdy people using a tiny ham radio.

Hosted by Professor Varun Gadkari