# Aligning 2D NMR Spectra Part 1

**R on Chemometrics & Spectroscopy using R**, and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)

Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

In one-dimensional \(^1\)H NMR spectroscopy, particularly biomolecular NMR, it is frequently necessary to align spectra before chemometric or metabolomics analysis. Poor alignment arises largely from pH and ionic strength induced shifts in aqueous samples. There are a number of published alignment algorithms for the one-dimensional case.

In this series of posts, I’ll discuss the alignment process for the case of 2D NMR, as implemented in the package `ChemoSpec2D`

.

No doubt you’ve compared two spectra by overlaying them on the screen, or printing them out, placing them on top of each other, and holding them up to the light. Conceptually, one can “align” spectra by a similar method: just slide one of the pieces of paper up/down and left/right until the spectra are optimally aligned. But how would one do this algorithmically? A literature searched turned up only a few publications on this topic. Among these, there was only one that I felt I could implement using the description in the paper: the HATS-PR algorithm of Robinette *et al.* (Robinette et al. 2011).

We’ll discuss the HATS algorithm in a future post. As a first step however, we need to consider how we know when two spectra are properly aligned. Visual inspection won’t work, as we will encounter cases where peaks in one region align, but only at the expense of peaks in another region. How would we rank such cases? To automate this process, we need to use an *objective function*, basically some kind of equation, that we evaluate as we explore the alignment space. A simple but effective option is to compute the distance between the two spectra. This is done by concatenating each row of the 2D spectra to give a long vector of intensities. The distance between these vectors can then be computed using any of the standard distance definitions. Let’s illustrate, starting by taking a look at some mis-aligned data. `ChemoSpec2D`

contains a mis-aligned data set, `MUD2`

, for just this purpose. Here are two spectra from `MUD2`

; note we are using the new convenience functions `LofC`

and `LofL`

to make it easy to overlay the spectra.

`library("ChemoSpec2D")`

`## Loading required package: ChemoSpecUtils`

```
data(MUD2)
mylvls <- seq(0, 30, length.out = 10)
plotSpectra2D(MUD2, which = c(1, 6), showGrid = TRUE,
lvls = LofL(mylvls, 2),
cols = LofC(c("red", "black"), 2, length(mylvls), 2),
main = "MUD2 Spectra 1 & 6")
```

The function `sampleDist`

allows us to compute the distance between every pair of spectra in the `MUD2`

data set, and present the results as a heat map. Here are the results using cosine as the distance measure.

```
cos_dist <- sampleDist(MUD2, method = "cosine",
main = "Cosine Distance")
```

The actual numerical values are in `cos_dist`

, a matrix. Looking at the heatmap, there are some modest differences among the spectra. However, if one looks at the scale, cosine distances are only defined on [-1 … 1]. While the cosine distance is popular in some spectroscopic contexts, it’s not the best objective function for our purpose because there is little absolute difference between -1 and 1 (and for `MUD2`

the absolute differences are even smaller, as the range is only 0, 0.67).

Let’s consider instead the Euclidean distance.

```
eu_dist <- sampleDist(MUD2, method = "euclidean",
main = "Euclidean Distance")
```

It turns of that the Euclidean distance gives a wider span of distances, which will serve us well in the next steps. Here, the range is roughly 80, 150. Note that in this plot the distance between identical spectra is zero, plotted as a white squares along the diagonal. When we used cosine as the distance, identical spectra were perfectly correlated and hence the diagonal in that plot was red.

In the next post I’ll discuss the general flow of the HATS-PR algorithm, and how to carry it out using `ChemoSpec2D`

.

# References

Robinette, Steven L., Ramadan Ajredini, Hasan Rasheed, Abdulrahman Zeinomar, Frank C. Schroeder, Aaron T. Dossey, and Arthur S. Edison. 2011. “Hierarchical Alignment and Full Resolution Pattern Recognition of 2D Nmr Spectra: Application to Nematode Chemical Ecology.” *Analytical Chemistry* 83 (5): 1649–57. https://doi.org/10.1021/ac102724x.

**leave a comment**for the author, please follow the link and comment on their blog:

**R on Chemometrics & Spectroscopy using R**.

R-bloggers.com offers

**daily e-mail updates**about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.

Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.