It’s a twenty-first century research conundrum: Suppose you are a scientist studying the microbiome in the human gut. You theorize that some of those bacteria are responsible for a health outcome, such as obesity or chronic lung disease. Using sophisticated technology, you gather a staggering array of information about the billions of bacteria that colonize the body, taken from a huge population of test subjects. But how do you analyze that data to accurately identify the responsible microbes, hiding among all the other organisms?
Luckily, you can call Martin T. Wells, Social Statistics/Computational Biology/Statistics and Data Science/Biostatistics in Healthcare Policy and Research.
Wells works with numbers, formulas, and gigabytes of very complex, high-dimensional data. Along with creating new theoretical frameworks, underlying data analysis methodologies, Wells also collaborates with researchers in a host of disciplines to apply the ideas to real-world problems.
Because current methods for analyzing data can’t keep pace with the amount of data scientists are generating, statisticians are in demand. “Data today has become so complex, so big, statisticians have to develop new methods of analysis and then immediately apply them in the field,” Wells says.
Analyzing Antibiotics: Why Do Certain Antibiotics Work the Way They Do?
Recently Wells joined with Kyu Y. Rhee, Medicine, Weill Cornell Medicine, to analyze the actions of certain antibiotics. Historically, antibiotic development depended on researchers’ understanding of phenotype-based readouts of bacterial growth inhibition or death. Over time, chemical genomic approaches helped to increase the understanding of antibiotic activity but still lacked the ability to differentiate compound activity.
Rhee and colleagues turned to metabolomics, the youngest of systems-level disciplines, which provides a global understanding of biological systems by studying their small molecule metabolites. The researchers applied metabolomic readouts in order to track the intrabacterial pharmacokinetic fates and pharmacodynamic actions of chemical compounds.
“It’s like working on a crime scene,” says Wells with relish. “The antibiotic kills the microbe, and Rhee gathers evidence of the ‘kill’ by looking at the metabolites that are active after the drug has done its job.”
Wells explains that metabolites are the enzyme products that fuel the biochemical activities of all cells. Certain metabolites indicate specific drug actions, which then inform researchers about bacterial growth inhibition or death. Growing evidence indicates that metabolites serve biological systems in equally specific qualitative and quantitative ways. Some drugs kill by suppressing an enzyme that is crucial to the bacteria and others by degrading the bacterial cell walls.
“There’s no way to do an experiment like this with pipettes and petri dishes in a lab…It’s all numerical. You can’t reach certain conclusions without looking at this big data. Computers and statistical methods are the new microscopes in modern biology.”
“That’s the ‘evidence’ of the crime scene. And we wanted to see if there was any pattern to that evidence from a metabolomics perspective,” says Wells. The researchers were trying to ascertain why certain antibiotics work the way they do. From that information, they hope to understand what characteristics of the target bacteria generate resistance to an antibiotic, which could lead to the design of more effective drugs.
With graduate student Irina Gaynanova and colleague James Booth, Biological Statistics and Computational Biology, Wells formulated a method to take the metabolomic data Rhee had generated and fit it into a statistical model to address the scientific problem. The analysis showed some unexpected results that may change scientists’ understanding of the biological processes behind antibiotics. “There’s no way to do an experiment like this with pipettes and petri dishes in a lab,” says Wells. “It’s all numerical. You can’t reach certain conclusions without looking at this big data. Computers and statistical methods are the new microscopes in modern biology.”
Diagnosing Tuberculosis: How Can Scientists Diagnose TB Earlier?
In another project at Weill Cornell Medicine, Wells and Rhee, along with Daniel W. Fitzgerald, Medicine, and Flonza Isa, an infectious disease fellow, searched for a biomarker for tuberculosis (TB) detection. TB is diagnosed currently through smear microscopy, cultures, and lung scans, but the disease has to be developed sufficiently in order to identify it. The Weill researchers wondered if it might be possible to identify TB earlier, using modern mass spectrometry-based metabolomic technologies to find new, specific molecular biomarkers in patients’ urine.
In an attempt to find telltale compounds, the Weill researchers ran mass spectrum analyses of urine from both TB-infected subjects and matched healthy controls. The mass spectrum data pinpointed every chemical compound present in the urine for each test subject, ultimately generating huge amounts of data that was then given to Wells.
Wells applied a methodology that he and Booth, along with PhD student Muting Wan developed to analyze mass spectrum data. He found an unknown compound in the urine of patients with TB. “This could be the basis of a whole new point-of-care test to diagnose TB early in the field,” says Wells. “If it works, it could be quite significant.”
The collaborative nature and ever-changing focus of Wells’ research means he’s constantly contributing cutting-edge findings to new fields of study. He has applied his expertise to research projects in biology, chemistry, biochemistry, genetics and other life sciences, as well as social sciences, law, and forensics. He’s an elected member of the Cornell Law School faculty as a result of his contribution to empirical legal studies, a cutting-edge area of law studies begun at Cornell in the 1980s by Theodore Eisenberg.
As part of his legal studies research, in 2012 Wells joined with Cornell Law School colleagues to write a series of papers on racial biases and other inequalities in Delaware’s application of the death penalty. Since, people pushing to abolish the death penalty in that state have cited the papers extensively.
More recently, Wells has talked with forensic experts about the challenges of crime scene data. He says that the huge scope and complexity of the data warrants statistical, principled methods of analysis.
“It’s a great time to be a statistician,” Wells concludes with satisfaction. “Our students are getting jobs in places we never got jobs before, such as Amazon, Google, and Yahoo! Research. When I started out as a statistician, we didn’t do very much with immediate application. I did a lot of somewhat abstract theory, which was interesting, but it had little impact. Now I’m applying these theoretical ideas to making a contribution to real-world problems. It’s a good feeling.”