A New Paradigm for Mass Spectral Identifications

There is a growing school of thought among practitioners that what is considered to be the prevailing existing paradigm for mass spectral identification is lacking. Glen Jackson, Professor of Forensic and Investigative Science at West Virginia University, addressed this need with his presentation "Expert Algorithm for Substance Identification (EASI): A New Paradigm for Mass Spectral Identifications" at SciX 2021, where it earned the FACSS Innovations Award. These awards are given for the most innovative and outstanding new research advancements debuted orally at the SciX Conference. Jackson spoke to Current Trends in Mass Spectrometry about this presentation and filling the aforementioned need.

Your presentation won the FACSS Innovation Award at the 2021 SciX Conference this past fall. Why did you see a need for a new paradigm?

The prevailing paradigm for all existing algorithms has been that the confidence in the identification of a questioned sample should increase as the spectral similarity of the questioned sample approaches that of a reference sample. This intuitive philosophy works fine when questioned and reference samples are collected close in time and on the same instrument. However, uncertainty grows, and error rates increase if questioned and reference samples are collected either at disparate times or using different conditions or instruments. Although it seems counterintuitive, our approach—our new paradigm—is that we do not require spectral similarity between a questioned spectrum and an existing reference spectrum for us to correctly identify a substance.

Your paradigm of choice was an expert algorithm for substance identification (EASI). How does your approach fill the need for better identification methods using MS?

Our approach fills two important needs of the forensic science community: 1) the ability to identify substances on an instrument without the need to acquire a reference sample on the same instrument, and 2) the ability to provide high confidence and low error rates with qualitative determinations like drug identifications.

Briefly summarize your new paradigm based on applying EASI and its implications.

EASI employs some long-standing and intuitive statistical tools to take advantage of the fundamental properties of electron ionization (EI) mass spectra. For example, the principles of EASI rely on a rigorous, mathematical understanding of statistical mechanics, transition state theory, and unimolecular fragmentation kinetics, which have been successfully applied to electron ionization mass spectra since the 1950s. These theories reveal that small (<20%) changes in excitation energy or effective observation times of ions from an EI source will cause the relative abundance of many pairs of ions in a spectrum to correlate in an approximately linear fashion. In other words, when the internal energy deposition surreptitiously increases during a day’s analyses, the relative abundance of some ions will increase relative to the base peaks and other ions will decrease relative to the base peak. Therefore, the daily, weekly, or monthly variance observed in replicate spectra of a substance will contain linear relationships in addition to random variance. These linear relationships are the key to EASI; we can model, and therefore compensate, for the predicted linear relationships.

We model the linear relationships using multivariate linear modeling, which was exquisitely detailed by Sir Francis Galton in the 1880s. Back then, he used multivariate linear modeling to predict the adult deviate height of human offspring from the deviate heights of their biological parents. Here, we use the abundance of major ions in a spectrum to predict the abundance of a held-out ion in the spectrum. The major benefit of linear modeling—in addition to its ease of understanding—is that linear equations can be interpolated between, or extrapolated beyond, existing data to enable accurate predictions in regions of observational space that are outside the bounds of the existing data. In other words, we can make predictions about ion abundances in spectra of a substance collected on other instruments in other laboratories, just by acquiring replicate spectra of a substance on one instrument in one laboratory. The spectra on the two different instruments do not have to be similar for the predictions to be accurate.

Regarding applications, we developed EASI specifically for the purpose of identifying seized drugs in a crime laboratory setting. We understand that “identification” from EI-MS data is only one part of a laboratory’s analytical scheme, but we wanted to maximize the informative power of EI-MS to a laboratory’s scheme. However, because the kinetics of unimolecular fragmentation apply to almost any molecule for which replicate fragmentation spectra can be acquired, EASI should be applicable to a wide variety of compounds on a wide variety of instruments. As an example, we originally developed EASI using replicate GC-MS data of cocaine and its diastereomers—like allococaine and pseudococaine—from more than 50 laboratories. We recently tested EASI on replicate tandem mass spectra of protonated precursors collected on an electrospray ionization tandem mass spectrometer (ESI-MS/MS), and the modeling explained more than 80% of the variance in the replicate spectra. We can therefore predict ion abundances, and discriminate between substances, that have remarkably similar structures/spectra in a wide variety of MS applications.

What were the biggest challenges encountered in developing EASI?

Gosh, there have been so many challenges. Honestly, it’s taken several years to co-develop the statistical tools and our understanding of the underlying statistical mechanics to the point where we now have confidence that they are well suited for one another. It has taken time to acquire large, relevant data sets and to prove that our application of the various statistical tools are appropriate and superior to conventional methods. Finally, I suspect that it will be challenging for the community to accept these ideas and incorporate EASI into casework or research. It would help if we could develop an easy graphical user interface (GUI) to automate its use.

Your group at West Virginia University’s long-term goal is to catalyze the progress of biomedical, analytical, and forensic research through the development of mass spectrometric instruments and techniques. Are there any other analytical applications where the EASI algorithm would be useful?

Absolutely. The most obvious and important applications of EASI will be applications that require the confident identification of known substances and the discrimination of those substances from spectrally similar isomers or compounds. Examples could include drug isomers, hydrocarbons, metabolites, lipids, peptides, oligosaccharides, and synthetic polymers.

Your presentation included input from associates J. Tyler Davidson (currently an assistant professor at Sam Houston State University, Huntsville, TX) and Samantha Mehnert (currently a PhD student at Purdue University, West Lafayette, IN). How did their expertise enhance the results of your research?

Tyler and Sam were pivotal in acquiring data, building databases, and testing some of the early statistical models. They were great listeners and effective critics; they helped filter out many bad ideas early on. Sometimes, just hearing a student trying to describe an idea back to me, or to another student, is helpful in developing a more effective way of communicating the idea.

What are the next steps regarding your work? Do you plan on applying your method for additional forensic drug-related applications, or to expand its use for different applications, or both?

Well, we haven’t yet published in the peer-reviewed literature, so that’s an obvious first step. We are grateful that our work is now funded by the National Institute of Justice (NIJ), and we are currently collaborating with several crime laboratories to build a database that contains more than 57,000 spectra of more than 70 fentanyl analogs. We have plenty of data analysis to perform on that dataset to establish whether or not EASI outperforms existing algorithms in its ability to correctly identify fentanyl analogs. Our first tests of EASI on distinguishing valerylfentanyl from isovalerylfentanyl and all other fentanyl analogs are proving to be considerably more effective than existing algorithms, so the near-term prospects are very exciting.

Glen Jackson is the Ming Hsieh Distinguished Professor of Forensic and Investigative Science at West Virginia University, where he also holds a joint appointment in the C. Eugene Bennett Department of Chemistry. Dr. Jackson earned a BS degree in the UK and MS and PhD degrees in the US, all in the area of analytical chemistry. His research is broadly defined as forensic and biological applications of mass spectrometry. His group's research has appeared in more than 90 publications, more than 160 conference and university presentations, and three issued patents. Since 2016, Dr. Jackson has served as the Co-Founder and Co-Editor-In-Chief of the journal “Forensic Chemistry.” He recently served a three-year term on the NIST OSAC subcommittee on Seized Drugs, and he has taught numerous workshops to practicing forensic professionals. He is an active forensic chemistry consultant, and his work has appeared on “Nancy Grace Live,” “Forensic Files II,” and “Law and Order SVU.” More information on the Jackson group can be found on his website, https://glenjackson.faculty.wvu.edu/.