Signatures for Mass Spectrometry Data Quality

Journal Article · Mon Mar 10 00:00:00 EDT 2014 · Journal of Proteome Research

DOI:https://doi.org/10.1021/pr401143e· OSTI ID:1129280

Amidan, Brett G. ^[1]; Orton, Daniel J. ^[1]; Lamarche, Brian L. ^[1]; Monroe, Matthew E. ^[1]; Moore, Ronald J. ^[1]; Venzin, Alexander M. ^[1]; Smith, Richard D. ^[1]; Sego, Landon H. ^[1]; Tardiff, Mark F. ^[1]; Payne, Samuel H. ^[1]

---

1. Pacific Northwest National Lab. (PNNL), Richland, WA (United States)

Ensuring data quality and proper instrument functionality is a prerequisite for scientific investigation. Manual validation for quality assurance is time consuming, expensive and subjective. Metrics for describing various features of LC-MS data have been developed to assist operators in discriminating poor (out of control) and good (in control) datasets. However, the wide variety of instrument specifications and LC-MS configurations precludes applying a simple range of acceptable values or cutoffs for such metrics. We explored a variety of statistical modeling approaches to predict the quality of LC-MS data. Using 1164 manually classified quality control (QC) LC-MS datasets, we fit logistic regression classification models to the QC data to predict whether a dataset is in or out of control. Model parameters were optimized by minimizing a loss function that accounts for the tradeoff between false positive and false negative errors. The optimal logistic regression classifier models detected bad data sets with high sensitivity (i.e. low false negative rate) while maintaining high specificity (i.e. controlling the false positive rate). As an example, predictions for Velos-Orbitrap instrumentation data had a sensitivity of 93.7% in detecting out of control datasets with a false positive rate of 8.3%. In comparison, we investigated the performance of several single metrics in predicting dataset quality. While maintaining a sensitivity of 93.7%, the corresponding false positive rates for these single-metric models unacceptably ranged from 32% to 97.7%. Finally, we evaluated the performance of the

• Research Organization: Pacific Northwest National Lab. (PNNL), Richland, WA (United States). Environmental Molecular Sciences Lab. (EMSL)
• Sponsoring Organization: USDOE
• DOE Contract Number: AC05-76RL01830
• OSTI ID: 1129280
• Report Number(s): PNNL-SA-97216; 47418; KP1601010
• Journal Information: Journal of Proteome Research, Vol. 13, Issue 4; ISSN 1535-3893
• Publisher: American Chemical Society (ACS)
• Country of Publication: United States
• Language: English

Related Subjects

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CHEMISTRY
Environmental Molecular Sciences Laboratory