ADVERTISEMENT
2 /3 FREE ARTICLES LEFT THIS MONTH Remaining
Chemistry matters. Join us to get the news you need.

If you have an ACS member number, please enter it here so we can link this account to your membership. (optional)

ACS values your privacy. By submitting your information, you are gaining access to C&EN and subscribing to our weekly newsletter. We use the information you provide to make your reading experience better, and we will never sell your data to third party members.

ENJOY UNLIMITED ACCES TO C&EN

ACS Meeting News

Attacking spectra to improve data analysis

In a bid to strengthen data handling, chemists prove that small changes to the baselines of spectra are enough to lead to errors in interpretation

by Celia Henry Arnaud
August 30, 2019 | APPEARED IN VOLUME 97, ISSUE 34

 

The reliance on spectroscopic data in various regulatory and legal settings can make it tempting to tamper with spectra to yield desired results. One possible way to tamper might be during the analysis of spectra, by making small perturbations in the data that can lead to major changes in interpretation. Garth Simpson and coworkers at Purdue University are using “adversarial spectroscopy” to figure out how to make data analysis approaches more resistant to such attacks, Simpson reported at the ACS national meeting in San Diego last week. The researchers made small changes to Raman spectra and evaluated how those changes affected the interpretation of the spectra. They used spectra of two polymorphs of the blood-thinning drug clopidogrel bisulfate as a test case. Linear discriminant analysis, a method that searches for data features that can separate things into categories, was able to easily separate the spectra of each polymorph and bare glass as belonging to separate classes. By making small, nearly imperceptible changes to the spectra, the researchers were able to nudge the system into misclassifying one polymorph, mistaking it for the other or the bare glass. The tiny changes the researchers made that were most effective in causing a misclassification weren’t to the dominant peaks in the spectra; they were to the baseline noise within the spectra. Simpson hopes that an awareness of the sensitivity of spectra to such attacks will lead to strategies for building better, more robust classification algorithms.

X

Article:

This article has been sent to the following recipient:

Leave A Comment

*Required to comment