Research map / Large language models / Agents, code and math
NanoBaseLib: A Multi-Task Benchmark Dataset for Nanopore Sequencing
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Presents NanoBaseLib, a multi-task benchmark dataset integrating 16 public Nanopore datasets (30M+ reads) re-processed with a unified pipeline, with baselines on four tasks: base calling, polyA detection, segmentation/event alignment, and RNA modification detection.
Problem
Nanopore sequencing data analysis methods are inaccurate across tasks, and ML researchers lack a large, integrated, uniformly preprocessed multi-task benchmark because public data are scattered and require heavy bioinformatic preprocessing.
Approach
Re-process 16 public R9.4 Nanopore datasets from raw signal with one unified pipeline (Guppy basecalling, minimap2 mapping, polyA detection, eventalign) so that a single dataset supports four interdependent tasks, with a software package to load data and compute metrics.
Key results
- Dorado v0.5.3 achieved the best RNA base calling match rate (93.96% on hek293t_wt) and Guppy v6.0.1 the best DNA match rate (93.60% on NA12878); best accuracy approximately 93%.
- Tailfindr showed the best polyA detection rate across all seven test datasets, while Nanopolish polya varied widely.
- SegPore had the best segmentation/event alignment scores (e.g., avg log-likelihood -2.778 on hek293T_wt), and m6Anet had the best m6A detection AUC (0.868 on miCLIP ground truth).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use Guppy v6.0.1 instead of Guppy v4.5.4 | DNA base calling · NA12878 | M Align: 91.93 -> 93.60 | Helped |
| Use Dorado v0.5.3 instead of Guppy v6.0.1 | RNA base calling · hek293t_wt | M Align: 91.78 -> 93.96 | Helped |
| Use Tailfindr rather than Nanopolish polya or Dorado | PolyA detection · seven test datasets | polyA detection rate | Helped |
| Use SegPore eventalign rather than Nanopolish eventalign | Segmentation and event alignment · hek293T_wt | Avg. std (σ̂): 3.073 -> 2.736 | Helped |
| Use SegPore eventalign rather than Nanopolish eventalign | Segmentation and event alignment · hek293T_wt | Avg. log p (L̂): -2.871 -> -2.778 | Helped |
| Use m6Anet rather than SegPore | m6A detection · mESCs_eligos, miCLIP | ROC AUC: 0.844 -> 0.868 | Helped |
| Use m6Anet rather than SegPore | m6A detection · mESCs_eligos, miCLIP2 | ROC AUC: 0.831 -> 0.891 | Helped |
| Use m6Anet rather than SegPore | m6A detection · mESCs_eligos, MeRIP-seq | ROC AUC: 0.72 -> 0.758 | Helped |
Surprising findings
- The best basecallers reach only about 93% match rate, far from the ~99.9% target for sequencing accuracy.
Limitations
- All public datasets are based on R9.4/R9.4.1 chemistry; performance on R10 is not assessed.
- Ground truth for base calling is approximate because RNA modifications alter current signals.
- Benchmarks used only a subset of the 16 datasets.
Experimental setup
Datasets: ont_polya_standard eGFP_polyA_DNA eGFP_polyA_RNA lambda_phage NA12878 curlcake scBY4741_m5C scBY4741_hm5C scBY4741_pU hct116 hek293t_wt hek293t_ko
Benchmarks: NA12878 hek293t_wt ont_polya_standard curlcake_synth hek293t_ko hct116 mESCs_eligos
Metrics: match/mismatch/insertion/deletion rate polyA detection rate mean squared error of polyA length average std average log-likelihood AUC-ROC AUC-PR
Baselines: Guppy Bonito Dorado Causalcall Rodan Nanopolish polya Tailfindr Nanopolish eventalign Tombo resquiggle SegPore eventalign m6Anet Nanom6A
Builds on: Guppy Nanopolish Tombo SegPore minimap2
Closest papers by meaning
- BEACON: Benchmark for Comprehensive RNA Tasks and Language Models (NeurIPS 2024 · similarity 0.82)
- DNABERT-2: Efficient Foundation Model and Benchmark For Multi-Species Genomes (ICLR 2024 · similarity 0.82)
- NovoBench: Benchmarking Deep Learning-based \emph{De Novo} Sequencing Methods in Proteomics (NeurIPS 2024 · similarity 0.81)
- Bridging Sequence-Structure Alignment in RNA Foundation Models (AAAI 2025 · similarity 0.81)
- Hyperbolic Genome Embeddings (ICLR 2025 · similarity 0.81)
- WelQrate: Defining the Gold Standard in Small Molecule Drug Discovery Benchmarking (NeurIPS 2024 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.