As ‘omics research advances, separations data such as chromatograms and nuclear magnetic resonance (NMR) spectra are being used as direct inputs to machine-learning algorithms ranging from principal components analysis to deep-learning neural networks. Using the data in this way would be improved by an ability to compare data across a wide range of instruments and facilities, but efforts have been hampered by a lack of harmony among measurements. There is, therefore, a need for quality control for the processes producing chromatographic and spectral data, which denotes the need for suitable reference materials (RMs). Presently, certified RMs contain reference data only on their composition or other physical properties. If data processing will involve spectral data, RMs should be certified for spectral profiles, and there is not yet a broadly accepted methodology for doing so. In this talk, I will present the results of recent NIST efforts in harmonizing metabolomic and proteomic measurements using data from interlaboratory studies. These efforts have resulted in a software package, interlab_py, which applies a mathematical procedure for automatically comparing spectra, assigning scores and determining outliers in populations of similar spectra, and identifying facilities which may have consistent process problems. The use of this procedure on data from two interlaboratory studies will be discussed, one using synthetic, biologically relevant metabolite mixtures and one using the NIST monoclonal antibody.
1. Understand the need for reproducible measurements in spectral and chromatographic data
2. Understand the basic principles behind the interlab_py software and its application to interlaboratory comparison of spectral data