Normalizing Gas‐Chromatography–Mass Spectrometry Data: Method Choice can Alter Biological Inference

Abstract

We demonstrate how different normalization techniques in GC-MS analysis impart unique properties to the data, influencing any biological inference. Using simulations, and empirical data, we compare the most commonly used techniques (Total Sum Normalization 'TSN'; Median Normalization 'MN'; Probabilistic Quotient Normalization 'PQN'; Internal Standard Normalization 'ISN'; External Standard Normalization 'ESN'; and a compositional data approach 'CODA'). When differences between biological classes are pronounced, ESN and ISN provides good results, but are less reliable for more subtly differentiated groups. MN, TSN, and CODA approaches produced variable results dependent on the structure of the data, and are prone to false positive biomarker identification. In contrast, PQN exhibits the lowest false positive rate, though with occasionally poor model performance. Because ESN requires extensive pre-planning, and offers only mixed reliability, and ISN, TSN, MN, and CODA approaches are prone to introducing artefactual differences, we recommend the use of PQN in GC-MS research.

Characteristics of Dietary Fatty Acids Isolated from Historic Dental Calculus of the 17th- and 18th-Century Inhabitants of the Subcarpathian Region (Poland)

“…The Agilent Technologies Mass Hunter version 07.06 (Palo Alto, CA, USA) was used as software for acquisition and quantification of the analysis data. For relative quantification of results, normalization method was applied [ 19 ]. Recovery during trimethylsilyl (TMS) derivatization step is from 90% to 106%, and precision %RSD is less than 4% [ 20 ].…”

Oviductal Extracellular Vesicles Improve Post-Thaw Sperm Function in Red Wolves and Cheetahs

“…Because protein expression levels were determined via UPLC-MS/MS, small differences in sample concentrations and/or measurement error can introduce artefactual differences across samples. To control for this effect, we used Probabilistic Quotient Normalization (PQN) [109], which has been shown to have the best performance of all routinely used normalization methods [110]. After PQN transformation, we filtered out proteins with fewer than 100 counts per million (1018 proteins).…”

Knowing Me, Knowing You: Anal Gland Secretion of European Badgers (Meles meles) Codes for Individuality, Sex and Social Group Membership

“…Because GC-MS profiles carry relative, and not absolute, information, even small differences in sample volume, concentration, in combination with impurities and background noise, can influence the apparent abundance of compounds, i.e., the 'size effect' (Noonan et al 2018). Because of this size effect, pre-processing must be carried out before profiles can be compared statistically.…”

“…We used Probabilistic Quotient Normalization (PQN; Dieterle et al 2006), which controls for the size effect by calibrating all profiles against the median profile (i.e., median peak values over all samples). Unlike other normalization methods, statistical analyses on PQN transformed data tend to have low false-positive rates, and can accurately recover groups of interest without introducing artefactual differences (Noonan et al 2018).…”

“…These analyses were conducted using the R package "randomForest" (RColorBrewer and Liaw 2018). We chose RF modeling as it does not require any parameter reduction prior to analysis (Cutler et al 2007), and has been shown to provide reliable results for biomarker identification (Noonan et al 2018). Identification of biomarkers important for classifying groups of interest in each RF model was carried out by applying a k-means clustering algorithm across RF variable importance values, and peaks in the cluster with the greatest mean variable importance were classified as a 'peak of primary importance' (PPI; Noonan et al 2018).…”