noisyR: enhancing biological signal in sequencing datasets by characterizing random technical noise

Abstract

High-throughput sequencing enables an unprecedented resolution in transcript quantification, at the cost of magnifying the impact of technical noise. The consistent reduction of random background noise to capture functionally meaningful biological signals is still challenging. Intrinsic sequencing variability introducing low-level expression variations can obscure patterns in downstream analyses. We introduce noisyR, a comprehensive noise filter to assess the variation in signal distribution and achieve an optimal information-consistency across replicates and samples; this selection also facilitates meaningful pattern recognition outside the background-noise range. noisyR is applicable to count matrices and sequencing data; it outputs sample-specific signal/noise thresholds and filtered expression matrices. We exemplify the effects of minimizing technical noise on several datasets, across various sequencing assays: coding, non-coding RNAs and interactions, at bulk and single-cell level. An immediate consequence of filtering out noise is the convergence of predictions (differential-expression calls, enrichment analyses and inference of gene regulatory networks) across different approaches.

bulkAnalyseR: An accessible, interactive pipeline for analysing and sharing bulk multi-modal sequencing data

“…The metadata table contains additional experimental information and represents the starting point for the DE analyses/comparisons. The pre-processing step (preprocessExpressionMatrix function) handles the countsbased noise detection using noisyR [13], which outputs a denoised, un-normalised expression matrix (Fig. S2).…”

“…Visualisation tools such as Cytoscape [11] facilitate hypothesis generation, but require external definitions of the networks. Moreover, GRN inferences are often performed on the entire dataset, generating complex, difficult-to-interpret networks [12] and lacking in robustness due to intrinsic variability [13]. GeNeCK [14] attempts, in a web-based setting, to link GRN inference and interpretation, but is decoupled from other steps such as DE analysis.…”

“…bulkAnalyseR enables the analysis of single- and multi-omics bulk-sequencing data, in an interactive Shiny interface, with several customisable components, from noise detection [13] to the identification of patterns [15] and GRN prediction [4]; the latter accommodates cis-, trans- and custom interactions. The setup facilitates a seamless transition to publication-ready figures without additional bioinformatics support.…”

ClustAssess: tools for assessing the robustness of single-cell clustering

“…The user can provide a feature set, including frequently used options such as the most abundant genes [49, 41], or highly variable genes [50, 16], or other custom subset specified by the user [57]. Also assessed are different sizes (number of entries) of the feature set; this incremental approach is essential for determining the largest subset of genes unaffected by noise [22, 21] which can alter downstream interpretations. The stability of each feature set is summarised using boxplots; the distribution of the ECC on the reduced UMAP space provides additional insights on the localisation of unstable regions on the UMAP topography.…”

“…While it is desirable to cluster cells based on biological signal, partitions can also be driven by technical nuisance effects (sequencing depth or noise [20, 21, 22]). Beyond the interplay of technical [23] and biological signals affecting clustering outputs, the results also depend on algorithmic decisions e.g.…”