m6Anet Analysis - Usage Guide

Overview

End-to-end m6A detection from Oxford Nanopore direct-RNA-sequencing (ONT DRS) signal data using m6Anet. Covers the required upstream pipeline (Dorado / Guppy basecalling -> minimap2 transcriptome alignment -> nanopolish eventalign -> m6anet dataprep -> m6anet inference), interpretation of the three per-site output columns (probability_modified posterior, mod_ratio stoichiometry, n_reads coverage) and the per-read modification probability output, the DRACH-only modeling constraint, minimum-coverage thresholds (20-50 reads per site), multi-condition comparison via xPore / Nanocompore / ELIGOS / Dorado native modification calling, reference-transcriptome version pinning, the cDNA-vs-DRS chemistry distinction, and orthogonal validation against MeRIP / GLORI.

Prerequisites

pip install m6anet ont-fast5-api xpore nanocompore
conda install -c bioconda minimap2 samtools nanopolish
# Dorado: download binary from https://github.com/nanoporetech/dorado

Reference inputs:

• POD5 / FAST5 raw signal files from a direct-RNA-sequencing run (kit SQK-RNA002 or SQK-RNA004; NOT cDNA-Nanopore)
• Transcriptome FASTA (GENCODE / Ensembl transcripts.fa; version pinned per project)
• Dorado modification model name + version (e.g., m6A_DRACH@v1); pin per project

Quick Start

• "Run the full m6Anet pipeline from POD5 to per-site m6A probabilities"
• "Filter m6Anet results for high-confidence sites (n_reads >= 20, prob >= 0.9)"
• "Compare m6A between two conditions with xPore Bayesian diffmod"
• "Run Nanocompore for modification-agnostic comparison between WT and KO"
• "First-pass m6A discovery with Dorado native modification calling on RNA004"
• "Cross-validate my m6Anet calls against published GLORI sites"

Example Prompts

Pipeline

"Basecall RNA004 POD5 files with Dorado rna004_130bps_sup@v5.0.0; align to GENCODE v44 transcriptome with minimap2 -ax map-ont -uf -k14 --secondary=no; run nanopolish eventalign with the m6Anet-required flag set; run m6anet dataprep and inference."

"Run m6anet dataprep and inference on a pre-computed eventalign.txt file from a HEK293T DRS run."

Filtering

"Filter m6anet results/data.site_proba.csv for high-confidence m6A sites: n_reads >= 20 AND probability_modified >= 0.9; export top 20 most-modified transcripts."

"Compute per-read modification rate per site by aggregating per-read CSV with threshold 0.5; report alongside per-site probability."

Differential Modification

"Run xPore diffmod between WT and METTL3-KO with 3 replicates per condition; report diff_mod_rate >= 0.1 AND pval < 0.05."

"Run Nanocompore sampcomp between two conditions on the same transcriptome; primary statistic GMM_logit_pvalue < 0.05."

Dorado Native

"Use Dorado modification calling on RNA004 data with m6A_DRACH@v1 model; report per-read modification probabilities for first-pass discovery."

"Cross-check Dorado high-recall site set against m6Anet for high-precision filtering."

Cross-Validation

"Compare my m6Anet site calls against published GLORI HEK293T m6A sites at gene level."

"Cross-reference m6Anet calls with MeRIP exomePeak2 peaks at common transcripts; report concordance."

Sanity Check

"Confirm my sequencing run is DRS (SQK-RNA002 / SQK-RNA004) and NOT cDNA-Nanopore; modification detection only works on DRS."

What the Agent Will Do

1. Verify DRS protocol (kit SQK-RNA002 / SQK-RNA004) before any modification-detection pipeline
2. Basecall POD5 / FAST5 with Dorado (modern) or Guppy (legacy); pin basecaller + model version
3. Align to TRANSCRIPTOME (not genome) with minimap2 -ax map-ont -uf -k14 --secondary=no
4. Sort and index BAM with samtools
5. Run nanopolish eventalign with the m6Anet-required flag pair --scale-events --signal-index (plus --summary and --threads for housekeeping; --samples and --print-read-names are needed only by other downstream tools)
6. Run m6anet dataprep + inference
7. Apply minimum-coverage filter (n_reads >= 20 conservative; >= 50 stringent)
8. Apply probability filter (probability_modified >= 0.9 high-confidence; >= 0.7 discovery)
9. For stoichiometry estimates, compute per-read modification rate
10. For differential between conditions, run xPore (Bayesian, no matched IVT needed) or Nanocompore
11. For RNA004 first-pass screening, run Dorado native modification calling alongside m6Anet
12. Cross-validate top sites against published GLORI / m6A-Atlas / MeRIP peaks
13. Report per-site probability, per-read modification rate, coverage, and orthogonal validation status

Tips

• m6Anet is DRACH-only by design. Non-DRACH sites are invisible. For non-DRACH discovery, use Dorado native or xPore / Nanocompore.
• cDNA-Nanopore data CANNOT be used for modification detection. PCR erases the signal. Only DRS (SQK-RNA002 / SQK-RNA004) works.
• RNA002 and RNA004 use different chemistry; models do NOT transfer. Pin chemistry version per project.
• Minimum coverage is non-negotiable. Sites with <20 reads at high probability are likely false positives.
• Per-site probability_modified is the model's posterior; per-read modification rate is the stoichiometry estimate.
• minimap2 for ONT DRS: use -ax map-ont -uf -k14 --secondary=no against TRANSCRIPTOME. Genome alignment breaks the m6Anet pipeline.
• nanopolish eventalign needs --scale-events --signal-index for m6Anet; --samples --print-read-names are needed only by other downstream tools (yanocomp, f5c-pipeline interop).
• m6Anet v1 used hyphenated CLI (m6anet-dataprep); v2 uses subcommands (m6anet dataprep). Pin version.
• Dorado modification models are versioned independently of the basecaller. Pin m6A_DRACH@vX explicitly.
• For absolute stoichiometry claims at named loci, cross-validate with GLORI (Liu 2023). Per-read modification rate has 5-15% error.
• Zhong 2023 Nat Commun 14:1906 benchmark documents tool tradeoffs; CHEUI is the only tool covering m6A AND m5C simultaneously.

Related Skills

• merip-preprocessing - MeRIP fragment-level cross-validation
• m6a-peak-calling - MeRIP peak set for orthogonal comparison
• m6a-differential - MeRIP differential at the loci where m6Anet has direct stoichiometry
• modification-visualization - Metagene and browser-track rendering of m6Anet site calls
• long-read-sequencing/basecalling - Dorado / Guppy basecalling fundamentals
• long-read-sequencing/long-read-alignment - General minimap2 alignment patterns
• long-read-sequencing/long-read-qc - DRS yield, length distribution, basecall accuracy
• long-read-sequencing/nanopore-methylation - DNA methylation calling sibling
• variant-calling/vcf-basics - Per-site variant call framework analogue
• rna-quantification/featurecounts-counting - For site-vs-MeRIP-peak comparison