Best Practices for Raman Spectroscopy Data Analysis

Data Analytics

Oct 13, 2021

Using Raman spectroscopy in biomanufacturing is an effective way to apply Process Analytical Technology (PAT) and monitor bioreactor analyte concentrations, like glucose, lactate, glutamate, and glutamine in a sample. PAT tools, like Raman, can become even more effective when the analyzers themselves are integrated within the bioreactor system. Integrated PAT makes sampling, sample analysis, data collection, and consolidation an automated process, making walk-away Raman spectroscopy analysis a reality.

^{This article is posted on our Science Snippets Blog}

When it comes to the best practices of Raman data analysis, we suggest Multivariate Data Analysis (MVDA). MVDA is a powerful tool to analyze data from dozen or hundreds of batch runs, and build powerful and robust prediction models based on Raman measurements.

Several Sartorius cell culture and bioreactor systems now have integrated Raman spectroscopy measurement via the universal BioPAT^® Spectro (read more)

Advantages of PAT for Bioreactor Monitoring and Control

Process Analytical Technology (PAT) tools such as Raman spectroscopy in combination with MVDA provide the capability to characterize and predict various analytes such as glucose, lactate amino acids. PAT can also be used to control a cell culture process and gain an understanding of cell metabolism, nutrient consumption, and quality.

PAT tools, such as Raman spectroscopy and MVDA, allow scientists to create more effective and tightly controlled bioprocesses with more frequent or continuous monitoring compared to classical analytical techniques. When a process is well controlled, it improves safety, efficacy, and robustness, and enables a Quality by Design, QbD approach.

Using Raman Spectroscopy to Control Critical Process Parameters

Raman spectroscopy is based on the inelastic scattering of photons from a monochromatic light source, usually from a laser source. When the photons interact with the molecules in a sample, the photons are scattered at a frequency higher or lower than the incident frequency. This is called the Raman effect and can be used to predict the quantities of certain kinds of molecules in the system.

Raman spectrometers collect huge amounts of information in the form of spectra. Based on the standalone spectra it is not possible to measure the concentration of the target molecules. This requires a reference analysis of the target molecules, such as glucose, lactate, or other metabolites. MVDA can be used to find correlations between the spectral data and the reference analysis data. Based on that a calibration model can be created to predict the concentration of the respective analytes of new samples.

Developing A Control Model

Raman-based process control relies on generating good quality data to develop robust calibration models that correlate spectral signals with reference data. This sort of model building requires the use of statistical techniques such as Design of Experiments (DOE) and MVDA to create predictive models that can be accurately scaled up.

Building accurate process models using data analytics can reduce the length of time needed for commercial development with fewer production runs and without waiting for out-of-specification (OoS) batches to occur. An accurate model can be created from Raman spectroscopy data and data analytics using SIMCA^® MVDA software and MODDE^® DOE software, which reduces the need for a time-intensive sample and laboratory analysis.

Difficult-to-measure analytes, which would previously have required a time- and cost-intensive sampling and laboratory analysis, can also be measured using Raman and MVDA. The models created provide the basis for a control strategy that can be transferred to larger-scale manufacturing platforms.

Overcoming Challenges in Model Building

Manufacturing of a biopharmaceutical product, such as a mammalian or microbial cell culture process, consists of an upstream and a downstream process. The upstream process is basically the cell culture process in the production bioreactor, which is followed by the downstream process of purifying the final product.

The current methods used to build a robust predictive control model require measuring multiple analytes over a course of repeated experiments to obtain statistically relevant data. When the data is being collected manually from commercial-scale bioreactors, this can be time-consuming and costly in terms of media, reagents, and staff time.

Using non-invasive PAT tools such as Raman spectroscopy, hundreds of data points for each sample and time point can be collected without stopping or interfering with the experiments. Furthermore, complementing this with a DOE approach ensures a level of statistical relevance that supports a robust prediction model, which is specific for the target analyte and does not depend on indirect correlations.

Using automated mini bioreactors also allows for a more cost-effective experimental design. Thus, the design space can be much larger than one provided by running hundreds of batches with random variations. A planned DOE can shorten the time frame and reduce the number of experimental runs needed to generate statically relevant data to build a predictive model.

Why a DOE Approach Is Better

Using integrated Raman spectroscopy in a bioreactor allows bioprocess scientists to collect spectral data in a more efficient way than using in-process variations from manufacturing bioreactors. Using spectral data with DOE makes it easier to:

Define a design space with more variation than would be expected for commercial use
Incorporate a range of typical and atypical process parameter values in the model-building data set
Specify “golden batch” runs with “standard” parameter settings that help to identify inevitable process variations during production cycles (Figure 1a)
Include parameter variations that cause varying process trajectories (Figure 1b) and thus reduce the correlation between analyte trends
Understand analyte spiking to induce a step change in analyte concentrations for either a single analyte or a range of analytes. This reduces correlations to other analyte trends and the range of analyte concentrations can be increased to ease the effect of a specific analyte on the spectra.

How to Generate Good Raman Spectroscopy Data for Predictive Analyte Models

A sample DOE workflow for producing the optimum data to build predictive analyte models with Raman spectroscopy involves the following steps:

Evaluate cell culture process and select areas where variations are likely to occur
Set up Ambr^® system for Raman measurements
Create DOE based on process relevant factors, e.g. CPPs to obtain widespread in process trajectories
Design an analyte spiking regimen
Set-up spiked analyte sample plates
Run Ambr^® experiments in-parallel with and without spiking
Consult with your Raman vendor to determine the ideal Raman measurement settings for your individual process
Use the same measurement settings for the Raman spectra acquisition for all runs
Transfer reference analyte and Raman data into MVDA analysis tools.

Designing an Analyte Spiking Regimen

After creating a DOE set-up, a spiking regimen is needed to add analytes of known concentration to the cell culture sample before measuring with Raman spectroscopy. Spiking is done to extend the ranges of the calibration model, since multivariate calibration models cannot extrapolate a concentration beyond the calibrated concentration range.

But more importantly, spiking breaks correlations between analytes, which can cause cross-sensitivity that could interfere with model building. This helps to create a robust calibration model which is not depending on other analytes.

Download a Technical Note

Get a technical note that covers a step-by-step guide for Ambr^® experimental set-up and statistical design for Raman BioPAT^® Spectro integration for generating high-quality calibration datasets.
The technical note also covers evaluating a cell culture process, creating a DOE, and offers input for designing an analyte spiking regime, as well as pre-processing Raman spectral data for use with SIMCA^®.

Download Note Now