Addressing the Challenges of Process Raman Spectroscopy

In recent years, Raman spectroscopy has been applied to process monitoring and control applications in a wide range of application fields, including bioprocessing, pharmaceuticals, food, oil and gas, and oceanography. Brian Marquardt, cofounder and CEO of MarqMetrix, Inc., and director and senior principal engineer with the Center for Process Analysis and Control in the Applied Physics Laboratory at the University of Washington, has more than 15 years of experience with such applications and recently spoke with us about his research. Marquardt is the winner of the 2016 Charles Mann Award presented by FACSS. This interview is part of the series with the winners of awards to be presented at the SciX 2016 conference.

Your research in applying Raman spectroscopy to process applications spans many fields, ranging from industrial process analysis and marine measurements to bioprocess monitoring. What do you consider to be the most challenging process application for Raman at the moment? Are you currently working on any projects that would address that challenge?

From my perspective, biotech applications of Raman are currently the most challenging because of the complex biological matrices and their associated fluorescence. We are currently working with a near-field focusing Raman immersion fiber probe (BallProbe, MarqMetrix) to effectively sample these systems with Raman as the optical density and the composition changes during a bioprocess. To reduce the fluorescence impact in a bioprocess application we are using a combination of unique data analysis methods and some very cleverly designed sampling interfaces. These developments allow us to look at the small-molecule initiators or products formed in the bioprocess with reduced interference from the cells or organisms, thereby greatly improving our signal-to-background ratio. I feel that Raman is an excellent tool for use in understanding bioprocessing once we mitigate the challenges due to biofluorescence.

In a recent paper (1), you demonstrated that Raman spectroscopy could provide real-time validated quantitative monitoring of continuous flow reactors (CFRs) when used in tandem with multivariate modeling. What benefits does this system and method offer compared to others?

The ability to collect in near real time (1 s) a high-resolution, quantitative Raman spectrum of a molecule in flow has many inherent benefits for process monitoring and quality control. We have demonstrated that the application of Raman in continuous flow reactors has many benefits. We have been using Raman as a quality assurance tool for molecule/reaction development in CFRs. The information-rich data provided by in-line Raman allows us to quickly perform complex design of experiment (DoE) studies to optimize the reaction space and conditions for production of that molecule. We have routinely used Raman as a quality assurance tool for the measurement and control of final product quality for many CFR applications in both the pharmaceutical and chemical industries.

Has the pharmaceutical industry shown interest in adopting this approach?

Yes, there has been significant interest from the pharmaceutical industry in what we have been doing with mating continuous flow reactors and Raman spectroscopy for improved process and product quality control. We have received support for some of our work in this field from a few leading pharmaceutical companies as well as from the Food and Drug Administration and the Bill and Melinda Gates Foundation.

You also presented work on a new method to detect cosmic spikes in process Raman data featuring a multistage spike recognition algorithm based on tracking sharp changes of intensity in the time domain (2). How does this method differ from previous methods for detecting cosmic spikes in process Raman data?

There are many algorithms for detecting and removing cosmic spikes from Raman spectra. They all have significant limitations and tradeoffs, and are suitable for only a narrow range of applications or hardware. Our algorithm was designed to be universal, robust, and easy to use. It has only one variable parameter that might need to be changed and is capable of discriminating cosmic spikes from rapid intensity changes of real Raman peaks.

How did you develop the multistage spike recognition algorithm?

At the time I had a visiting student from Scotland (who later became an excellent post-doc). He had an idea and shared it with me. I found several gigabytes of process Raman data that were accumulated in my lab over a decade that would make a great test set. We sat together and worked through the data refining the algorithm, and eventually we developed a version of the algorithm that worked best with all the data sets. We had a great set of reproducible process Raman data to test the idea, and that was one of the reasons for our success.

In what situations do cosmic spikes cause the greatest difficulty in interpreting process Raman spectra?

Cosmic spikes are manageable in all practical circumstances, except for cases when measurement times are very long. We have had some challenges with cosmic spikes in applications such as Raman imaging and the determination of very low analyte concentrations that involved long integration times.

You have also been involved with fusing data from process Raman, infrared (IR), and nuclear magnetic resonance (NMR) analyses for crude oil fractions (3). What are the advantages provided by this approach?

The primary advantage for developing data fusion algorithms is to deal with the real-world process analysis challenge: “There is no perfect analyzer that can measure all of my process problems.” Therefore, we began to look at the systematic coupling of analytical tools that gave us a more complete view of the measurement challenges faced with complex processes or products. We began to couple complementary tools such as Raman and Fourier transform infrared (FT-IR) spectroscopy to fully describe the vibrational signatures of a product or process. We found that we could improve our models and better understand our process, but it was still not completely informative to describe all crude oil fractions. We required an orthogonal analytical technique to help describe variations in chemical structure of the samples that was not being captured by the vibrational methods. The addition and fusion of the NMR data into our models greatly reduced the classification and prediction errors in our system. 

What are the main challenges you and your group faced in this research?

The biggest challenges were determining how to combine data sets with greatly different intensity and wavelength scales and have them represent equal weight in a process control model. We had to develop numerous algorithms to deal with scaling and weighting of both spectral data (Raman, IR, NMR, mass spectrometry, UV–vis, and so on) and univariate data (temperature, pressure, refractive index, and so forth) to build robust and effective process models using data fused inputs.

Which of your research projects have yielded the greatest step forward in process Raman spectroscopy?

I think that the development of the Raman immersion ballprobe sample interface was a huge step forward for process Raman. The ability to have a sampling interface for Raman spectroscopy that was effective for measuring many different sample types including solids, slurries, powders, liquids, and gases reproducibly was a big step forward. A stable and reproducible sample interface facilitated the application of Raman in a variety of process applications from pharma to food to the deep ocean at high temperatures and pressures. Since then we have designed, built and optimized many different versions of the immersion probe that are making an impact across a number of industrial and environmental application fields (for example, biotech, oceanography, food, pharma, and oil and gas).

What are the next steps in your research?

We are working on several projects where high precision, accuracy, and long-term stability of process Raman measurements are very important. When you start chasing certain levels of these metrics, you have to start accounting for many intricate details about both the hardware and interface properties of the measurement system that can otherwise be ignored. In this scenario, it is crucial to have a deep understanding about all aspects of measurements (from hardware to optics and data processing) and have the right team of scientists and engineers in place to develop the correct approach to the challenge. We have truly enjoyed applying Raman to many exciting process applications and are currently working on automating our many of our new techniques, making them more generic and valuable at the same time.

References

• M.F. Roberto, T. I. Dearing, S. Martin, and B.J. Marquardt, J. Pharm. Innov. 7, 69–75, DOI 10.1007/s12247-012-9128-8 (2012).

• S. Mozharov, A. Nordon, D. Littlejohn, and B. Marquardt, Applied Spectroscopy66(11), 1326–1333, DOI: 10.1366/12-06660 (2012).

• T.I. Dearing, W.J. Thompson, C.E. Rechsteiner, Jr., and B.J. Marquardt, Applied Spectroscopy 65(2), 181–186, DOI: 10.1366/10-05974 (2011).