The JMP Development team has been hard at work, and they have implemented some great new tools for chemometric analysis capability into the software structure. Many thanks to @clay_barker, @RyanParker, and @chris_gotwalt1 for their superpowers that allowed them to pull all this chemometric analysis capability into one platform. Also, a special shout out to @MikeD_Anderson for his contributions to peak analysis and peak finding based on his Spectris Add-in, which can be found in the User Community File Exchange.
In this blog, @Mark_Bailey and I fill you in on what’s new in the latest release of JMP and JMP Pro 18. Several new direct models have been added to the FDE toolbox. These modeling types are better for direct interpretability, especially when analyzing mixtures.
There is also a new Python script that allows you to open .spc files, thanks to @nick_shelton and a new Python script from @ChristianStopp that allows you to open .spa files.
New Direct Functional PCA methods
JMP has added four new options to the existing Direct Functional PCA tab:
- Multivariate Curve Resolution (MCR)
- Unconstrained MCR
- Penalized Singular Value Decomposition (SVD)
- Penalized Nonnegative SVD
These new methods build on current capability and can be found in the Functional Data Explorer when you go to Models > Direct Models.
Multivariate Curve Resolution
Multivariate Curve Resolution (MCR) has been used for many years in the analysis of chemometric data. It is especially useful when analyzing spectra from mixtures and discerning the components of a mixture, so much so that MCR is also known as self-modeling mixture analysis.
Below are images of the NMR analysis of a mixture of alcohols: butanol, propanol, and pentanol. The data is set in a wide format for analysis, and there is no initial pre-processing or data cleaning. There are 23 rows of data, and the mixture components are determined based on the NMR spectra of the pure alcohols.
The analysis takes around 30 seconds, so don’t worry if you don’t see something immediately-. Based on the above image, it took 20 shape functions or functional principal components to get the best fit with this method-. Overall, the model did well in identifying the breakdown of the mixture components.
MCR comes in two flavors: regular and Unconstrained MCR-. Unconstrained MCR allows for negative weighting in the shape functions. In a blog from Daniel Pelliccia, “Multivariate Curve Resolution: An Introduction”, Pelliccia states “that a non-negativity constraint only makes sense when you are sure it applies to the functions you are trying to estimate, as a way to speed up the convergence process. Get rid of it in the general case.”
Penalized Singular Value Decomposition (SVD)
There are two options for Penalized SVD: Penalized Non-negative SVD constricts the analysis to non-negative weighting like MCR, and Penalized SVD allows for negative weighting like Unconstrained MCR-. Both SVD options use a LASSO-type penalty on the loadings and the singular values to enable the zeroing out of inactive loadings.
MCR and SVD Comparison by Score Plots
By looking at the score plots of the Unconstrained MCR, MCR, and Penalized SVD shown below, the Unconstrained MCR had a more challenging time discerning the pure spectra for pentanol (1020) versus the other two models based on the sub-grouping of other nearby spectra. For reference, propanol is spectrum 1008, and butanol is spectrum 1018.
Accessing more data file types
With the release of JMP 18 and its improved Python integration, there are two new options for importing spectral data into JMP for analysis.
Both .spa and .spc files can now be imported directly into JMP thanks to Christian Stopp for .spa and Nick Shelton for .spc, respectively. Both file formats are based on different instrument packages, with .spa coming from OMNIC software and .spc from Grams software. Both file types are imported in a tall format and must be transposed for further preprocessing and analysis.
JMP is always looking to improve our software, and we hope you will let us know which tools would be helpful for analyzing chemometric data, especially spectra and chromatograms.