The local-balanced model for improved machine learning outcomes on mass spectrometry data sets and other instrumental data.

One unifying challenge when classifying biological samples with mass spectrometry data is overcoming the obstacle of sample-to-sample variability so that differences between groups, such as between a healthy set and a disease set, can be identified. Similarly, when the same sample is re-analyzed under identical conditions, instrument signals can fluctuate by more than 10%. This signal inconsistency imposes difficulties in identifying subtle differences across a set of samples, and it weakens the mass spectrometrist's ability to effectively leverage data in domains as diverse as proteomics, metabolomics, glycomics, and imaging. We selected challenging data sets in the fields of glycomics, mass spectrometry imaging, and bacterial typing to study the problem of within-group signal variability and adapted a 30-year-old statistical approach to address the problem. The solution, "local-balanced model," relies on using balanced subsets of training data to classify test samples. This analysis strategy was assessed on ESI-MS data of IgG-based glycopeptides and MALDI-MS imaging data of endogenous lipids, and MALDI-MS data of bacterial proteins. Two preliminary examples on non-mass spectrometry data sets are also included to show the potential generality of the method outside the field of MS analysis. We demonstrate that this approach is superior to simple normalization methods, generalizable to multiple mass spectrometry domains, and potentially appropriate in fields as diverse as physics and satellite imaging. In some cases, improvements in classification can be dramatic, with accuracy escalating from 60% with normalization alone to over 90% with the additional development described herein.

Clinically Viable Assay for Monitoring Uromodulin Glycosylation.

Uromodulin, also known as the Tamm-Horsfall protein or THP, is the most abundant protein excreted in human urine. It is associated with the progression of kidney diseases; therefore, changes in the glycosylation profile of this protein could serve as a potential biomarker for kidney health. The typical glycomics analysis approaches used to quantify uromodulin glycosylation involve time-consuming and tedious glycoprotein isolation and labeling steps, which limit their utility in clinical glycomics assays, where sample throughput is important. Herein, we introduce a radically simplified sample preparation workflow, with direct ESI-MS analysis, enabling the quantification of N-linked glycans that originate from uromodulin. The method omits any glycan labeling steps but includes steps to reduce the salt content of the samples, thereby minimizing ion suppression. The method is effective for quantifying subtle glycosylation differences of uromodulin samples derived from different biological states. As a proof of concept, glycosylation from samples that differ by pregnancy status were shown to be differentiable.

The Aristotle Classifier: Using the Whole Glycomic Profile To Indicate a Disease State.

"The totality is not, as it were, a mere heap, but the whole is something besides the parts."-Aristotle. We built a classifier that uses the totality of the glycomic profile, not restricted to a few glycoforms, to differentiate samples from two different sources. This approach, which relies on using thousands of features, is a radical departure from current strategies, where most of the glycomic profile is ignored in favor of selecting a few features, or even a single feature, meant to capture the differences in sample types. The classifier can be used to differentiate the source of the material; applicable sources may be different species of animals, different protein production methods, or, most importantly, different biological states (disease vs healthy). The classifier can be used on glycomic data in any form, including derivatized monosaccharides, intact glycans, or glycopeptides. It takes advantage of the fact that changing the source material can cause a change in the glycomic profile in many subtle ways: some glycoforms can be upregulated, some downregulated, some may appear unchanged, yet their proportion-with respect to other forms present-can be altered to a detectable degree. By classifying samples using the entirety of their glycan abundances, along with the glycans' relative proportions to each other, the "Aristotle Classifier" is more effective at capturing the underlying trends than standard classification procedures used in glycomics, including PCA (principal components analysis). It also outperforms workflows where a single, representative glycomic-based biomarker is used to classify samples. We describe the Aristotle Classifier and provide several examples of its utility for biomarker studies and other classification problems using glycomic data from several sources.