In pharmaceutical and other industries that rely on spectroscopy and multivariate calibration for quality control of manufacturing processes, optimizing the analysis of spectral data is imperative. Using a tool that is specifically designed with spectral analytics in mind can make the job faster, easier and more reliable.
Let’s take a look at one tool that does just that: the spectroscopy skin available for SIMCA multivariate data analytics software.
Optimizing spectral data analysis with tools designed especially for spectroscopy analytics can help keep multivariate calibration modeling running smoothly.
What is a spectroscopy skin?
A spectroscopy skin is a special tool offered for SIMCA software that can simplify the analysis of spectral data and aid in the development of reliable multivariate calibration models. It automatically applies a number of alternative settings and puts the necessary views and options for spectral data analysis quickly at your fingertips.
Consider these three reasons to use the SIMCA spectroscopy skin:
1. It’s quick. The spectroscopy skin gives you a customized interface with easier access to spectroscopy related tools, settings and plots. The most frequently used spectroscopy analysis tools are concentrated in one place.
2. It’s easy. At the heart of the skin is a wizard that makes it easy to access filters and compare a number of models at the same time. You can also import additional spectral data for new samples and apply the settings of your model to the new data with one click using the “import to predict” or “import to model” features.
3. It’s reliable. The spectroscopy skin uses a well-established and recognized technology and applies the most widely accepted MVDA standards for spectral analysis and multivariate calibration. The default settings used are ideally suited for spectral data. In addition, you’re able to compare multiple models at once for better predictive power.
How does it work?
Using the spectroscopy skin starts with a wizard to set up the basis for spectral filter comparison. This makes it easy for you to compare up to three spectral filters against the performance of the original, un-preprocessed data. The wizard will automatically establish four datasets and present the corresponding regression models.
To start, you have three filter options in the wizard that you select to compare against the default (raw data). These are: the SNV (standard normal variate correction) and the first and second derivative filters. You can run the wizard multiple times to study additional filters or combine them in a sequence. (Filters run in a sequence are called “chained”.)
The wizard also comes with support for new plots. You can plot spectra to visualize the impact of filters, and have two plots ready for rapid and transparent plot performance. These are: comparing the Q2 and RMECV (Root Mean Square Error of Cross Validation) across the different models in the same plot.
Select the spectral range
When going into the filter comparison wizard, the first view you’ll see is one called “select spectral range.” Your selection here will apply to all four models that you’ll be calculating. So first, you have to decide if you want to work with the entire spectral window, or if there are certain sections in the spectral window where you know or expect that there is less information and would like to exclude those.
The tool allows you to select the variables (columns) that should be included or excluded in your model.
Select spectra for model training
Next, you’ll see an option to select the spectra to be used in the model. Previously, you selected which columns to use, but here you’ll select which rows to include. The selection will be applied to all four models that are about to be calculated.
The next is the stage called “Specify Filters and y-variable.” You’ll select which of the available Y- variables you want to work with. If you do not specify any y-variables, the resulting models would be principal component analysis (PCA) models. If you specify a y-variable, you’ll have a choice between PLS and OPLS for the model. By default, the raw data will be included. You can add new filters or combine them into a sequence (chaining).
After that, you’ll come to the stage when the different filters are applied to the data so you have a visual representation of the impact of the filters that you selected.
In the top left, you will always see the original data (no filter) and in the other three plots, you will see the results of the different filters you have selected.
At this point, if you’re happy with the four filters you have selected, you can click “finish” and the four models you selected will be calculated. The filter data becomes new datasets.
Applying additional plots
After your datasets are created and your models are computed, you can apply additional wizards to view more plots, such as Q2, which allows you to compare the predictive power across the different models.
Using the Q2 option, you’ll see a new plot comparing model performance from different models. You can also view addition models with chained filters.
If you want to investigate the impact of more filtering tools, you can run the filter wizard multiple times. But keep in mind when you re-run the filter comparison wizard, the raw data will always be there, so it will be shown multiples times in your chained filter view.
Another plot that comes with the skin is the RMSECV (Root Mean Square Error of Cross Validation). Here you can look for models with the lowest RMSECV to be the preferred choice.
When you exit the wizard, you can access a button called the Plotting Spectra. This is a shortcut to align for spectral plots of 4 datasets.
You can plot not only the original spectra here, but also alternative pre-processings. You can also plot the loadings of the different models where you have the spectral information on the x-axis.
Two additional functionalities in the spectroscopy skin are the Import to Predict tool and Import to Model tool.
The Import to Predict tool allows you to assess and define the prediction set in one go. For more on this, watch the demo below. The demo also covers how to import and predict and compare RMSEP for multiple models. (Which is similar to the RMSECV plot).
The Import to Model functionality allows you to import data and update the current model in one step. The current model is the one you have highlighted in the project window. So you can add additional data and update the model to include additional samples. When you do that, the preprocessing for the selected model will automatically be applied to the spectra that you import.
Want to know more?
Watch the demo in this recorded webinar to understand more about how the spectroscopy skin in SIMCA works.
Download the presentation