pmartR: Quality Control and Statistics for Mass Spectrometry-Based Biological Data

Abstract

Prior to statistical analysis of mass spectrometry (MS) data, quality control (QC) of the identified biomolecule peak intensities is imperative for reducing process-based sources of variation and extreme biological outliers. Without this step, statistical results can be biased. Additionally, liquid chromatography–MS proteomics data present inherent challenges due to large amounts of missing data that require special consideration during statistical analysis. While a number of R packages exist to address these challenges individually, there is no single R package that addresses all of them. We present pmartR, an open-source R package, for QC (filtering and normalization), exploratory data analysis (EDA), visualization, and statistical analysis robust to missing data. Example analysis using proteomics data from a mouse study comparing smoke exposure to control demonstrates the core functionality of the package and highlights the capabilities for handling missing data. In particular, using a combined quantitative and qualitative statistical test, 19 proteins whose statistical significance would have been missed by a quantitative test alone were identified. The pmartR package provides a single software tool for QC, EDA, and statistical comparisons of MS data that is robust to missing data and includes numerous visualization capabilities.

Activity‐Based Protein Profiling of Chitin Catabolism

“…MSGF probability scores were used to calculate the FDR value (0.05 cut off). Peptides were filtered and normalized using the an approach adapted from Stratton et al (2019): [33] 1) redundant and contaminate human peptides were removed. 2) Peptides without two values in the entire dataset were removed.…”

Differential effects of redox conditions on the decomposition of litter and soil organic matter

“…Two additional samples from the high-frequency treatment were also excluded due to low number of peaks: one on day 36 160 and one on day 44. To identify potential outliers, we computed robust Mahalanobis distance (rMd) from peak abundance data using the 'rmd_filter' function in the R package 'pmartR' (Stratton et al, 2019). The rMd data were then mapped to P values using five metrics, including correlation coefficient, fraction of missing data, median absolute deviation, skewness, and kurtosis (Matzke et al, 2011).…”

What is Normalization? The Strategies Employed in Top-Down and Bottom-Up Proteome Analysis Workflows

“…Normalization of MS data can be approached in a number of ways depending on the experiment being performed and the overall aim of the analysis [81,85,86,87,88,89]. Experiments that focus on gaining as much coverage of the proteome as possible can benefit from methods that use “cross run” methods that seek to infer the validity of detected peptides by assessing the abundance of those peptides across an entire experiment [90].…”

“…For many workflows this will not be an issue. For instance, where samples are precious or in low abundance (such as micro-dissected human-derived tissue sections), there may simply not be enough material to use high stringency identification conditions reliably [85]. As discussed earlier, the overriding assumption is that the number of peptides injected into the mass spectrometer is also normalized, as are the sample processing conditions.…”

“…Therefore, it seems logical that if a particular sample contains an error of more than 1 to 2 standard deviations of the median, the sample should be removed from the cohort, and the analysis re-performed. There are also a host of algorithmic techniques that have also been shown to be effective [85,86,90,92] however a full exploration of the operation of the mathematics behind these algorithms is beyond the scope of this review. What can be concluded is that the large quantity and diversity of computational tools suggests that no single algorithm is effective for all experiment types.…”