Variable reference alignment: an improved peak alignment protocol for NMR spectral data with large intersample variation.

In an effort to address the variable correspondence problem across large sample cohorts common in metabolomic/metabonomic studies, we have developed a prealignment protocol that aims to generate spectral segments sharing a common target spectrum. Under the assumption that a single reference spectrum will not correctly represent all spectra of a data set, the goal of this approach is to perform local alignment corrections on spectral regions which share a common "most similar" spectrum. A natural beneficial outcome of this procedure is the automatic definition of spectral segments, a feature that is not common to all alignment methods. This protocol is shown to specifically improve the quality of alignment in (1)H NMR data sets exhibiting large intersample compositional variation (e.g., pH, ionic strength). As a proof-of-principle demonstration, we have utilized two recently developed alignment algorithms specific to NMR data, recursive segment-wise peak alignment and interval correlated shifting, and applied them to two data sets composed of 15 aqueous cell line extract and 20 human urine (1)H NMR profiles. Application of this protocol represents a fundamental shift from current alignment methodologies that seek to correct misalignments utilizing a single representative spectrum, with the added benefit that it can be appended to any alignment algorithm.