Skip to content

usage-guide.md

Latest commit

 

History

History
66 lines (49 loc) · 2.04 KB

usage-guide.md

File metadata and controls

66 lines (49 loc) · 2.04 KB

Medaka Polishing - Usage Guide

Overview

Medaka uses neural networks to polish consensus sequences and call variants from Oxford Nanopore data. Models are trained on specific basecaller versions for optimal accuracy.

Prerequisites

conda create -n medaka -c conda-forge -c bioconda medaka
conda activate medaka

Quick Start

Tell your AI agent what you want to do:

  • "Polish my draft assembly with medaka"
  • "Call variants from ONT reads using medaka"

Example Prompts

Assembly Polishing

"Polish my draft assembly draft.fa using reads.fastq.gz with medaka"

"Run medaka consensus on my Nanopore assembly using the R10.4.1 SUP model"

Variant Calling

"Call haploid variants from my ONT reads aligned to the reference"

Model Selection

"What medaka model should I use for R10.4.1 flowcell with SUP basecalling?"

"List available medaka models and help me choose the right one"

What the Agent Will Do

  1. Identify the correct medaka model based on your basecaller and chemistry
  2. Run medaka_consensus for polishing or medaka_variant for variant calling
  3. Handle intermediate files and cleanup
  4. Report improvement in consensus quality

Model Selection

Critical: Use the model matching your basecaller.

# List available models
medaka tools list_models

# Common choices:
# - r1041_e82_400bps_sup_v5.0.0 (R10.4.1, SUP, latest)
# - r941_min_sup_g507 (R9.4.1, SUP)

Expected Accuracy

Basecaller Before Polish After Polish
SUP R10.4.1 Q20 (~99%) Q40+ (>99.99%)
HAC R10.4.1 Q15-18 Q30+

Tips

  • Wrong model is the most common cause of poor polishing - always check basecaller version
  • For diploid variant calling, use Clair3 instead (medaka diploid is deprecated)
  • Reduce batch size (-b 50) if running out of memory
  • Process by region (--region chr1) for large genomes
  • Use GPU if available for much faster processing

Resources