Learning and Imputation for Mass-spec Bias Reduction (LIMBR)

Abstract

Motivation: Decreasing costs are making it feasible to perform time series proteomics and genomics experiments with more replicates and higher resolution than ever before. With more replicates and time points, proteome and genome-wide patterns of expression are more readily discernible. These larger experiments require more batches exacerbating batch effects and increasing the number of bias trends. In the case of proteomics, where methods frequently result in missing data this increasing scale is also decreasing the number of peptides observed in all samples. The sources of batch effects and missing data are incompletely understood necessitating novel techniques. Results: Here we show that by exploiting the structure of time series experiments, it is possible to accurately and reproducibly model and remove batch effects. We implement Learning and Imputation for Mass-spec Bias Reduction (LIMBR) software, which builds on previous block based models of batch effects and includes features specific to time series and circadian studies. To aid in the analysis of time series proteomics experiments, which are often plagued with missing data points, we also integrate an imputation system. By building LIMBR for imputation and time series tailored bias modeling into one straightforward software package, we expect that the quality and ease of large-scale proteomics and genomics time series experiments will be significantly increased.

Circadian Proteomic Analysis Uncovers Mechanisms of Post-Transcriptional Regulation in Metabolic Pathways

“…This represented 3,776 proteins that had peptides at every time point, and 976 proteins for which levels of some peptides missing <30% of time points were imputed using the K-nearest neighbors (KNN) algorithm within LIMBR (Crowell et al, 2018; see STAR Methods). The data were then subjected to modeling and removal of batch effects using a time-series-specific algorithm within LIMBR that improves rhythm recognition in the presence of mass spectrometric batch effects (Crowell et al, 2018; see STAR Methods). In brief, LIMBR modeled the replicate and time series correlations to produce a matrix of residuals containing the batch effects, which were then modeled to produce linearly independent bias trends.…”

“…To examine the role of CSP-1 in regulating the metabolic proteome, we sampled fungal mats from a Δ csp-1 strain and prepared samples for TMT-MS analysis as done for the wildtype strain. 4,742 proteins were detected and imputed in all samples using LIMBR (see STAR Methods) (Crowell et al, 2018). Analysis of these data by eJTK_cycle identified 1,316 proteins (~28% of the identified Δ csp-1 proteome) as circadianly rhythmic with p < 0.05 (Hutchison et al, 2015), a proportion similar to the fraction of rhythmic wild-type proteins (1,273 or 27%).…”

“…A potential caveat to this conclusion has been the small numbers of rhythmic proteins identified, often less than 1% of the proteome, reflecting technical and analytical issues. Our extensive replicate time course, use of recent technological advances (TMT-MS), and improved computational techniques (Crowell et al, 2018) allowed analysis of a larger proportion of the predicted proteome in our organism (45%), from which we found 27% of detected proteins significantly cycled in abundance on a daily basis. Greater than 40% of the significantly rhythmic proteome was not associated with a corresponding significantly rhythmic transcript (Figures 2A, S4A, and S4B) and the long average phase-delay (~10-hour) between peak mRNA and protein abundance (Figure 2B) both support the importance of circadian post-transcriptional regulation in not only controlling which proteins are rhythmic but also the time of day at which these rhythmic proteins peak (Schwanhäusser et al, 2011).…”

“…“pre-alpha” as of Oct. 3 2016 (Crowell et al, 2018). In this approach, a missing value of a peptide in a given sample is replaced by the average value in that sample of the K (10) other peptides which had the most similar expression across other samples for which the peptide with the missing value was observed; i.e.…”

“…Our sampling density, combined with computational tools developed to address issues of missing data and batch effects (Crowell et al, 2018), has allowed the identification of a significantly larger portion of the circadian proteome than previously achieved. Overall, the rhythmic proteome was highly enriched in metabolic functions, and rhythmically expressed enzymes seem to be coordinated, resulting in the generation of metabolic pathway “phases.” We found the correlation between rhythmic mRNA and rhythmic protein was only 60%, highlighting extensive post-transcriptional regulation.…”