Journal: Nucleic Acids Research

Article Title: Detecting a wide range of epitranscriptomic modifications using a nanopore-sequencing-based computational approach with 1D score-clustering

doi: 10.1093/nar/gkae1168

Figure Lengend Snippet: Average F1-scores of the five algorithms compared in this study on the E. coli and S. cerevisiae rRNA test dataset ( NC : Nanocompore; DRM : Drummer; E - DSE: Epinano Delta-Sum-Error; E - LR : Epinano Linear Regression). The E. coli and S. cerevisiae rRNA datasets comprise 10 independent samples. Each sample contains eight subsamples with coverage-depths ranging from 10 to 2000. Different coverage-depths were used since algorithm performance depends on the coverage-depth, as indicated by recent studies ( , ) and also confirmed by our results. Note that all positions are treated as either positive or negative since unsupervised algorithms, do not distinguish between different modification types. In line with this, we do not compute separate F1-scores for each modification type separately, but rather only one F1-score for the whole dataset (for the given coverage-depth). As shown, Modena outperformed other algorithms across all coverage-depths; in some cases by a large margin (e.g. at coverage-depths of 50, 75, 100 and 200). The performance of all algorithms was very stable across the 10 independent samples . Thus, although the figure above shows average F1-scores, the results are highly consistent across all Samples 1–10.

Article Snippet: The E. coli K12 MG1655 strain was grown in 1× LB Broth Miller (1st Base, Singapore) without antibiotics at 37°C at 160 rpm shaking.

Techniques: Modification

Journal: Nucleic Acids Research

Article Title: Detecting a wide range of epitranscriptomic modifications using a nanopore-sequencing-based computational approach with 1D score-clustering

doi: 10.1093/nar/gkae1168

Figure Lengend Snippet: Precision–Recall curves (PR curves) for Sample 1 ( E. coli and S. cerevisiae rRNA dataset) for different coverage-depths. As shown, resampling increases the area under the PR curves (i.e. AUPRC scores) across all coverage-depths. Kuiper test further improves AUPRC scores across all coverage-depths, although to a lesser extent.

Article Snippet: The E. coli K12 MG1655 strain was grown in 1× LB Broth Miller (1st Base, Singapore) without antibiotics at 37°C at 160 rpm shaking.

Techniques:

Journal: Nucleic Acids Research

Article Title: Detecting a wide range of epitranscriptomic modifications using a nanopore-sequencing-based computational approach with 1D score-clustering

doi: 10.1093/nar/gkae1168

Figure Lengend Snippet: Violin plots of Modena score distributions for positive and negative test cases across different coverage-depths for Sample 1 of the E. coli / S. cerevisiae benchmark dataset are shown. Two well-separated clusters can be seen for all coverage-depths. The final Step 5 of our algorithm (1D score-clustering) leverages this separation to determine the classification threshold. Note that this represents a different paradigm from the standardly used P -value based thresholds. As shown in our study, this approach is not limited to Modena and can, in principle, be applied to any threshold-based unsupervised algorithm.

Article Snippet: The E. coli K12 MG1655 strain was grown in 1× LB Broth Miller (1st Base, Singapore) without antibiotics at 37°C at 160 rpm shaking.

Techniques:

Journal: Nucleic Acids Research

Article Title: Detecting a wide range of epitranscriptomic modifications using a nanopore-sequencing-based computational approach with 1D score-clustering

doi: 10.1093/nar/gkae1168

Figure Lengend Snippet: Average F1-scores (for Samples 1 through 10, E.coli / S. cerevisiae dataset) with coverage-depths ranging from 10 to 2000 are shown. Drummer : original Drummer algorithm with P -value and odds ratio-based threshold; Drummer + 1D clustering : Drummer algorithm (i.e. G-test statistic) with 1D score-clustering step (see Figure ). For detailed results across all samples, see and .

Article Snippet: The E. coli K12 MG1655 strain was grown in 1× LB Broth Miller (1st Base, Singapore) without antibiotics at 37°C at 160 rpm shaking.

Techniques:

Journal: Nucleic Acids Research

Article Title: Detecting a wide range of epitranscriptomic modifications using a nanopore-sequencing-based computational approach with 1D score-clustering

doi: 10.1093/nar/gkae1168

Figure Lengend Snippet: Average F1-scores (for Samples 1 through 10, E.coli / S. cerevisiae dataset) with coverage-depths ranging from 10 to 2000 are depicted. Epinano: Epinano-DSE algorithm with z-score based threshold; Epinano + 1D clustering : Epinano-DSE algorithm with 1D score-clustering step (see Figure ). For detailed results across all samples, see and .

Article Snippet: The E. coli K12 MG1655 strain was grown in 1× LB Broth Miller (1st Base, Singapore) without antibiotics at 37°C at 160 rpm shaking.

Techniques: