Journal: Genome Biology

Article Title: Campolina: a deep neural framework for accurate segmentation of nanopore signals

doi: 10.1186/s13059-026-03950-1

Figure Lengend Snippet: Segmentation quality evaluation. a Predicted (C) and ground-truth (R) border positions are compared using Jaccard similarity (J), its expanded variant allowing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pm 1$$\end{document} ± 1 positions ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text {J}_{\textrm{exp}}$$\end{document} J exp ), and the bi-directional Chamfer distance (CD). Event-level alignment is performed by index-overlap matching (MS), constructing an alignment matrix and traceback to identify matches (m), insertions (i), and deletions (d). Alignment ratios and an alignment score (AS) are computed from aligned pairs (A). Event accuracy is quantified by the L1 distance between z-normalized event means ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _C$$\end{document} μ C , \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _R$$\end{document} μ R ). Finally, reference k-mers are assigned to predicted events, and Pearson’s r is computed between predicted event means ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _C$$\end{document} μ C ) and expected k-mer levels ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _k$$\end{document} μ k ). b The segmentation quality evaluation for R10.4.1 Zymo D6322 dataset, R10.4.1 Human NA12878 dataset, R9.4.1 Zymo D6322 dataset, and R9.4.1 Human NA12878 dataset, in that order. Both Zymo datasets are prepared without E. coli reads that were used for training. Segmentation results are compared against the segmentation obtained by the Scrappie algorithm with the hyperparameter set defined by Sigmoni (denoted as Scrappie S) and RawHash2 (denoted as Scrappie R). All metrics are calculated as shown in ( a ). L1 and Pearson’s r are calculated for all alignments and matches only

Article Snippet: Finally, reference k-mers are assigned to predicted events, and Pearson’s r is computed between predicted event means ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _C$$\end{document} μ C ) and expected k-mer levels ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _k$$\end{document} μ k ). b The segmentation quality evaluation for R10.4.1 Zymo D6322 dataset, R10.4.1 Human NA12878 dataset, R9.4.1 Zymo D6322 dataset, and R9.4.1 Human NA12878 dataset, in that order.

Techniques: Variant Assay

Journal: Genome Biology

Article Title: Campolina: a deep neural framework for accurate segmentation of nanopore signals

doi: 10.1186/s13059-026-03950-1

Figure Lengend Snippet: Segmentation suitability evaluation for R10.4.1 nanopore version. Performance is evaluated on three tasks: Zymo binary classification, Host depletion, and Zymo multiclass classification. Metrics reported include length-weighted accuracy, precision, recall, F1 score, and unclassified rate. Unclassified rate is the percentage of reads that could not be assigned to any of the provided references in the RawHash2 framework. We treat those reads as misclassified when calculating other metrics, but also provide the details on the “unclassified rate” to quantify the ratio of reads that cannot be aligned to any reference. Sigmoni classifies all reads; therefore, we do not report the “unclassified” rate when evaluating segmentation in the Sigmoni framework. Both segmentation strategies (Scrappie and Campolina) are evaluated using two classification frameworks (Sigmoni and RawHash2) and three classification tasks

Article Snippet: Finally, reference k-mers are assigned to predicted events, and Pearson’s r is computed between predicted event means ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _C$$\end{document} μ C ) and expected k-mer levels ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _k$$\end{document} μ k ). b The segmentation quality evaluation for R10.4.1 Zymo D6322 dataset, R10.4.1 Human NA12878 dataset, R9.4.1 Zymo D6322 dataset, and R9.4.1 Human NA12878 dataset, in that order.

Techniques:

Journal: Genome Biology

Article Title: Campolina: a deep neural framework for accurate segmentation of nanopore signals

doi: 10.1186/s13059-026-03950-1

Figure Lengend Snippet: Evaluation of read mapping. The results for mapping the Zymo dataset to the reference database containing Zymo references (Zymo to Zymo), and the Zymo and human dataset to the reference database containing Zymo references and CHM13 reference (Zymo+human to Zymo+CHM13). Evaluation is performed for R9.4.1 (denoted R9) and R10.4.1 (denoted R10) nanopore versions. Metrics report precision, recall, and F1 score calculated with UNCALLED pafstats

Article Snippet: Finally, reference k-mers are assigned to predicted events, and Pearson’s r is computed between predicted event means ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _C$$\end{document} μ C ) and expected k-mer levels ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _k$$\end{document} μ k ). b The segmentation quality evaluation for R10.4.1 Zymo D6322 dataset, R10.4.1 Human NA12878 dataset, R9.4.1 Zymo D6322 dataset, and R9.4.1 Human NA12878 dataset, in that order.

Techniques:

Journal: Data in Brief

Article Title: A Sentinel-1 SAR imagery dataset for airstrips detection and segmentation in the Brazilian Amazon Rainforest

doi: 10.1016/j.dib.2026.112472

Figure Lengend Snippet: Data representation and annotation formats. (a) Example optical image (not in dataset). (b) Corresponding Sentinel-1 SARmimage. (c) Binary segmentation mask. (d) Bounding box annotation overlaid on the SAR image for the object detection task. Scale bar in (a) represents 1000 m and applies to all images. North is up.

Article Snippet: Mendeley Data S1-AAD: Sentinel-1 Amazon Airstrip Dataset (Original data)

Techniques:

Journal: PeerJ

Article Title: JGR-NMF: joint graph-regularized non-negative matrix factorization for spatial domain identification

doi: 10.7717/peerj.20585

Figure Lengend Snippet: (A) Manually annotated ground truth labels. (B) Bar chart comparing ARI, NMI, and PUR of different methods on the Breast Cancer-1 dataset. (C) Spatial domain identification visualizations.

Article Snippet: Please see the access instructions in the to access the following datasets: The Breast Cancer-1 Dataset is available at: https://www.10xgenomics.com/datasets/human-breast-cancer-block-a-section-1-1-standard-1-1-0 .

Techniques:

Journal: Genome Biology

Article Title: Campolina: a deep neural framework for accurate segmentation of nanopore signals

doi: 10.1186/s13059-026-03950-1

Figure Lengend Snippet: Segmentation quality evaluation. a Predicted (C) and ground-truth (R) border positions are compared using Jaccard similarity (J), its expanded variant allowing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pm 1$$\end{document} ± 1 positions ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text {J}_{\textrm{exp}}$$\end{document} J exp ), and the bi-directional Chamfer distance (CD). Event-level alignment is performed by index-overlap matching (MS), constructing an alignment matrix and traceback to identify matches (m), insertions (i), and deletions (d). Alignment ratios and an alignment score (AS) are computed from aligned pairs (A). Event accuracy is quantified by the L1 distance between z-normalized event means ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _C$$\end{document} μ C , \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _R$$\end{document} μ R ). Finally, reference k-mers are assigned to predicted events, and Pearson’s r is computed between predicted event means ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _C$$\end{document} μ C ) and expected k-mer levels ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _k$$\end{document} μ k ). b The segmentation quality evaluation for R10.4.1 Zymo D6322 dataset, R10.4.1 Human NA12878 dataset, R9.4.1 Zymo D6322 dataset, and R9.4.1 Human NA12878 dataset, in that order. Both Zymo datasets are prepared without E. coli reads that were used for training. Segmentation results are compared against the segmentation obtained by the Scrappie algorithm with the hyperparameter set defined by Sigmoni (denoted as Scrappie S) and RawHash2 (denoted as Scrappie R). All metrics are calculated as shown in ( a ). L1 and Pearson’s r are calculated for all alignments and matches only

Article Snippet: Finally, reference k-mers are assigned to predicted events, and Pearson’s r is computed between predicted event means ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _C$$\end{document} μ C ) and expected k-mer levels ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _k$$\end{document} μ k ). b The segmentation quality evaluation for R10.4.1 Zymo D6322 dataset, R10.4.1 Human NA12878 dataset, R9.4.1 Zymo D6322 dataset, and R9.4.1 Human NA12878 dataset, in that order.

Techniques: Variant Assay

Journal: Genome Biology

Article Title: Campolina: a deep neural framework for accurate segmentation of nanopore signals

doi: 10.1186/s13059-026-03950-1

Figure Lengend Snippet: Segmentation suitability evaluation for R9.4.1 nanopore version. Performance is evaluated on three tasks: Zymo binary classification, Host depletion, and Zymo multiclass classification. Metrics reported include length-weighted accuracy, precision, recall, F1 score, and unclassified rate. Unclassified rate is the percentage of reads that could not be assigned to any of the provided references in the RawHash2 framework. We treat those reads as misclassified when calculating other metrics, but also provide the details on the “unclassified rate” to quantify the ratio of reads that cannot be aligned to any reference. Sigmoni classifies all reads; therefore, we do not report the “unclassified” rate when evaluating segmentation in the Sigmoni framework. Both segmentation strategies (Scrappie and Campolina) are evaluated using two classification frameworks (Sigmoni and RawHash2) and three classification tasks

Article Snippet: Finally, reference k-mers are assigned to predicted events, and Pearson’s r is computed between predicted event means ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _C$$\end{document} μ C ) and expected k-mer levels ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _k$$\end{document} μ k ). b The segmentation quality evaluation for R10.4.1 Zymo D6322 dataset, R10.4.1 Human NA12878 dataset, R9.4.1 Zymo D6322 dataset, and R9.4.1 Human NA12878 dataset, in that order.

Techniques:

Journal: Genome Biology

Article Title: Campolina: a deep neural framework for accurate segmentation of nanopore signals

doi: 10.1186/s13059-026-03950-1

Figure Lengend Snippet: Evaluation of read mapping. The results for mapping the Zymo dataset to the reference database containing Zymo references (Zymo to Zymo), and the Zymo and human dataset to the reference database containing Zymo references and CHM13 reference (Zymo+human to Zymo+CHM13). Evaluation is performed for R9.4.1 (denoted R9) and R10.4.1 (denoted R10) nanopore versions. Metrics report precision, recall, and F1 score calculated with UNCALLED pafstats

Article Snippet: Finally, reference k-mers are assigned to predicted events, and Pearson’s r is computed between predicted event means ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _C$$\end{document} μ C ) and expected k-mer levels ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _k$$\end{document} μ k ). b The segmentation quality evaluation for R10.4.1 Zymo D6322 dataset, R10.4.1 Human NA12878 dataset, R9.4.1 Zymo D6322 dataset, and R9.4.1 Human NA12878 dataset, in that order.

Techniques: