COMPANY CONFIDENTIAL. DO NOT DISCLOSE.
Cracking the case with data-driven forensics
Analytically-minded forensic chemists streamline the testing and development of state-of-the-art new technologies to help bring criminals to justice.
_____________________________________________________________________________________________
Challenge: Develop new tools and methodologies to help forensic scientists uncover key evidence that solves crimes.
Solution: Leverage JMP for design of experiments (DOE) methods such as factorials and response surfaces that optimize multiple experimental conditions simultaneously.
Results: With JMP, researchers have the experimental capabilities and advanced analytics they need to gain quantitative insights into new forensic tools and advance the development of new forensic methodologies. Says researcher Ashur Rael: “JMP broadens the options for how our research might progress. We can ask questions and find answers that we couldn’t find otherwise.”
_____________________________________________________________________________________________
In the now familiar crime drama trope, detectives play a starring role in cracking even the toughest case. But it’s often forensic scientists back in the lab who uncover the evidence that turns the tide and solves the crime. In this context, art imitates life; forensic chemistry has become a central part of the criminal justice system. And as key players in a field that revolves around proof beyond the shadow of a doubt, forensic scientists strive to continually improve their analytical methods.
With that goal in mind, researchers at Indiana University – Purdue University Indianapolis (IUPUI) have turned to JMP® to streamline the experimentation process.
At a recent JMP Discovery Summit, Ashur Rael, a graduate student in IUPUI’s Department of Chemistry and Chemical Biology, a paper entitled “Improving Gas Chromatography–Vacuum Ultraviolet Spectroscopy (GC–VUV) for Forensic Science Using JMP,” written in collaboration with co-authors Associate Professor John Goodpaster, graduate assistants Courtney Cruse and Zackery Roberson and undergraduate assistant Magnus Rydberg, all at IUPUI.
In his presentation, Rael detailed how he and his research collaborators apply clustering and multivariate analysis in JMP to uncover underlying spectral characteristics that reveal differentiation and quantify GC–VUV’s ability to identify compounds. In the process, they’re helping to develop new tools and methodologies that will aid forensic scientists in solving crimes faster and with scientifically-sound evidence to prove it.
Knowns and unknowns
Forensic chemistry applies the disciplines of chemistry and toxicology to identify unknown materials involved in crimes. Analyzing samples of controlled substances, ignitable liquids or explosives requires instruments and methodologies with high levels of sensitivity, selectivity and specificity.
One such methodology is GC–VUV. This relatively new approach combines gas chromatography (for the separation of gas mixtures) with vacuum ultraviolet spectroscopy (for capturing quantitative and qualitative spectral data on gaseous compounds).
Rael is applying JMP to improve method development and spectral analysis for GC-VUV. Using JMP for design of experiments (DOE) applications, including factorials and response surfaces, enables him to optimize multiple experimental conditions simultaneously. It also allows him to consider multiple responses.
“Developing new methods by testing one factor at a time is time-consuming and difficult, especially when multiple responses are important,” Rael says. “Understanding the spectral data is difficult or impossible using visual or direct analysis.”
The Goodpaster research group is now looking at the identification of individual molecules within a sample and creating overall material profiles. “We might be identifying a specific drug within a sample or fibers that can be matched to a known. I created a hierarchical clustering for identification of unknowns, and for identification and comparison between knowns and unknowns,” Rael explains.
Smaller than the Large Hadron Collider
GC-VUV is conducted on an instrument originally developed for the petrochemicals industry, in answer to industry experts’ need to analyze molecules that are highly similar to one another. The device is a detector for chromatography, which separates mixtures by passing them through a medium in which the mixture components move at different rates.
“Say you have a glass of orange-water that includes various compounds from oranges, plus residual organics from the water,” Rael says. “If you have good chromatographic separation, you can see each piece individually. To achieve that, in this case we’re looking at the interaction of very high-energy light coming off the molecules.”
In the past, such analysis required a sizable synchrotron facility – the biggest example of which is the Large Hadron Collider, operated by the European Organization for Nuclear Research (CERN) in a 17-mile tunnel near Geneva. The instrument Rael works with is much more manageable – about the size of a carry-on suitcase.
The results have been promising. Currently, the optimal detection method is based on mass. But if the same atoms are arranged in different ways to form unique molecules, it’s impossible to use mass to tell the molecules apart. In forensics, for example, it can be very hard to distinguish among “designer drugs” – synthetic analogs of legally-restricted narcotics, which are devised to circumvent drug laws.
With the new detection method, Rael builds upon the fact that every molecule has its own signature. “This could become the first choice for a lot of analysis that’s currently impractical,” he says.
Tiny differences, huge datasets
But performing this kind of spectral analysis can involve very large sets of data. “Let’s say we’re looking at phenylethylamines, a class of drug that includes methamphetamines,” Rael says. “Because they’re similar molecules, they interact with light in similar patterns.”
In fact, an interaction point occurs every .05 nanometers. “We’re talking about hundreds of data points in each spectrum,” Rael points out. “And it’s hard to readily differentiate them. It’s not like a 15-character name with only a few permutations. It’s more like a 100,000-character name.”
What’s more, Rael is working with infinitesimally small quantities of each substance – parts per billion in a single microliter. “We’re talking about nanograms,” he adds. Rael looks for a correlation coefficient with a library of existing spectra – essentially the space between two spectral lines of the experimental value and the library value.
“With nearly 1,000 data points per spectrum, you can’t typically put that directly into discriminate analysis, because you need to have fewer dimensions than samples,” Rael explains. “So we start with a principal components analysis to do dimensionality reducing. Then we can start visualizing something.”
Becoming the first choice
That’s where JMP comes in. “JMP has very nice features, excellent usability and excellent data outputs,” Rael says. “We can do dimensionality reductions. We can do discriminate analysis. And then we end up with score plots. So it’s telling us about the classifications.
“Without JMP, principal component analysis is largely qualitative, not quantitative,” he continues. “We can determine the average of distances between groups, but it becomes unwieldy, because then we’re doing one-to-one comparisons, and we end up with vectors. So we don’t get digestible information.”
With JMP, Rael and his collaborators explore average prediction probabilities to gain a more quantitative value. He can go beyond simpler questions of classification to drill down into more useful details. Rael cites the example of petroleum compounds: “In arson cases, you look for characteristic molecules of an accelerant like fuel,” he says. “Then you can narrow down the types of fuels.”
If Rael can see that everything in his DOE design space for the molecules has a probability of one, then he can verify there’s no uncertainty in the GC-VUV results. With mass spectrometry, in contrast, every molecule could have uncertainty.
“We look for the weaknesses in mass spectrometry, and we look for the weaknesses in vacuum-ultraviolet spectroscopy,” Rael says. “Because you can’t replace half of the picture with a different half of the picture. If vacuum-ultraviolet spectroscopy can tell you more of the story with minimal compromises, then it could become the first choice.”
JMP on the scene
Part of the challenge, though, is convincing the forensic community of the advantages of a new methodology like GC-VUV. “That’s why a lot of our analysis goes into illustrating how different the spectra are, to eliminate that uncertainty,” Rael explains.
In addition to using JMP for spectral comparison, the IUPUI research team applies the software to method development, including using design of experiments to improve methods and develop new ones. With this, the group aims at increasing the reproducibility of their scientific outputs.
Rael is also combatting a “one-thing-at-a-time” mindset in the forensics community. “We usually have to start with a more robust design,” he says. “So we’ll start with a full factorial or a response surface. And then we’ll do iterative optimizations within that space to get to higher resolutions.”
That provides similar data to doing one factor at a time, Rael notes, “because we’re gaining sub-screening designs, for example. But because we’re structuring our experimental design into these design-of-experiment spaces, we can look at the interactions among the various factors.”
JMP has become crucial to these efforts, he says; “JMP is an awesome tool. It provides excellent exploratory analysis and I’m excited to look further into its graphing tools.
“JMP broadens the options for how our research might progress,” Rael concludes. “We can ask questions and find answers that we couldn’t find otherwise.” And, if all goes well, support the creation of new tools and methodologies to help forensic scientists solve more cases.
