Nanopore Adaptive Sampling For Targeted Sequencing

Adaptive sampling is a real time enrichment method for Oxford Nanopore sequencing that selectively sequences or rejects specific DNA molecules as they pass through a pore. By comparing the first few hundred bases of each read against a reference genome, the sequencer decides whether to continue sequencing the molecule or reverse the voltage and eject it. This guide is for molecular biologists, clinical researchers, and bioinformaticians who want to understand the practical decisions and implementation steps for targeted long read sequencing without upfront amplification or probe based capture.

Adaptive sampling works on the principle that each molecule entering a pore generates a short current signal. The basecaller identifies the first chunk of sequence and aligns it to a target reference Galaxy Training Network. If the read matches the target region, the pore continues sequencing the full molecule. If the read comes from an off target region, the sequencer flushes the molecule and opens the pore for a new candidate. The result is an enriched population of long reads spanning your loci of interest without the bias introduced by PCR or the hands on steps of hybridization capture.

At a Glance

Decision Criteria For Adaptive Sampling

You should consider adaptive sampling when your research question targets specific genomic loci but you also want the long read advantages of nanopore sequencing. The technique shines for projects where you need to detect structural variants, phase alleles, or assess base modifications across defined regions EMBL EBI Training. If you are validating a candidate pathogenic variant identified by short read sequencing, adaptive sampling lets you zoom into that locus with kilobase long reads and observe the surrounding context including repetitive elements and mobile element insertions.

The decision depends on three main factors. First, the size of your target region. Adaptive sampling works best when the total target space is between 0.5% and 10% of the genome. If your targets are too small, the reject rate is very high and the effective throughput drops. If your targets are too large, enrichment becomes negligible because most reads are accepted anyway. Second, the complexity of the genome. In highly repetitive genomes, the initial alignment may be ambiguous and the pore may waste time on reads it cannot confidently place. Third, the desired read length. Adaptive sampling enriches for longer reads because the decision window only captures the first few hundred bases. Short fragments may be rejected before they are properly classified, so you should size select your library to favor fragments above five kilobases.

For clinical applications where you need to detect large structural rearrangements or mobile element insertions that short reads miss, adaptive sampling has proven effective. One study identified a pathogenic SINE VNTR Alu insertion in the MSH2 gene that had been missed by standard sequencing panels Fam Cancer. Another group used adaptive sampling to simultaneously profile DNA methylation and sequence across imprinted regions, demonstrating that the method can combine genomic and epigenomic data from a single library Sci Rep.

Practical Workflow Or Implementation Sequence

The workflow for adaptive sampling has six major stages. Each stage requires specific attention to detail to avoid costly troubleshooting during the sequencing run.

1. Define the Target BED File

Start with a BED file that lists every chromosome, start coordinate, and end coordinate you want to enrich. You can include hundreds of separate intervals. The sequencer will treat any read that aligns to these coordinates as a target read. Include 500 to 1000 base pairs of flanking sequence on each side of your region of interest. This padding accounts for the alignment uncertainty during the real time decision. A common mistake is to design a target file with exact exon boundaries. The read may start upstream of the exon and be flagged as off target if the initial aligned chunk falls outside your padded interval.

2. Prepare High Molecular Weight DNA

Library quality is the single largest determinant of adaptive sampling success. You need long, intact DNA molecules. Shear your DNA gently and size select to retain fragments longer than five kilobases. The MinION flow cell pores operate more efficiently when molecules are long. Short molecules pass through the pore too quickly and the basecaller may not have enough data to make a confident alignment decision within the time window.

3. Load the MinKNOW Software Configuration

Open MinKNOW and choose the adaptive sampling option under the sequencing run settings. Upload your BED file and select the reference genome that matches the species you are sequencing. MinKNOW will build an index of the reference for real time alignment. Choose a decision time window. The default is 0.2 seconds after the start of translocation. If your target regions are unique in the genome, you can keep the default. If your genome is repetitive, increase the window to 0.4 seconds to give the aligner more data for accurate classification.

4. Start the Run and Monitor Enrichment

During the first 30 minutes, the sequencer calibrates its pore selection and the enrichment may appear low. After this stabilization period, monitor the adaptive sampling status in MinKNOW. The software displays the proportion of reads accepted and rejected in real time. You should see an enrichment factor of 5 to 30 fold depending on your target size. If the reject rate is very high, your target region may be too small. If the accept rate is close to 100%, your target region may be too large for meaningful enrichment.

5. Basecall With Adaptive Sampling Aware Parameters

After the run finishes, basecall the raw signal data with Guppy or Dorado using the high accuracy model. The adaptive sampling decision does not affect downstream basecalling. You can treat the enriched reads as standard nanopore data. Align them to your reference with a long read aligner such as minimap2. Extract the reads that overlap your target intervals for downstream analysis. You do not need to filter out off target reads because they were already rejected at the pore, but a small fraction of off target reads will still slip through.

6. Analyze Variants and Modifications

Use variant callers designed for long reads such as Clair3 or Medaka to identify small variants and structural variants. If you are interested in base modifications, use the Remora or megalodon pipelines to call 5mC or other epigenetic marks across your target regions Bioconductor. The long read context allows you to phase variants and modifications on the same haplotype, which is a major advantage over short read based epigenetic methods.

Quality Checks

Before you proceed with biological interpretation, verify that the enrichment worked as intended. Count the number of target reads in your aligned dataset and compare them to the number you would expect from a standard nanopore run. If your target region is 1% of the genome, you should see at least 5% to 10% of your reads mapping to the target. If the enrichment factor is below 3 fold, check whether your BED file padding is sufficient and whether your reference index is compatible with the version of MinKNOW you used.

Inspect the read length distribution. Adaptive sampling tends to reject shorter fragments more often, which can bias your structural variant calling if you rely on the full length distribution for detection. Compare the median read length in your target region to the median read length of the entire run. A large discrepancy may indicate that you are missing structural variants that reside on shorter fragments.

Use the real time log files from the run to assess the decision accuracy. MinKNOW writes a summary of accept and reject events. If more than 20% of accepted reads later align to off target regions, your alignment parameters during the run were too permissive. Adjust the decision time window or the alignment stringency in your next run.

Common Mistakes

The most frequent mistake is designing a target BED file without flanking padding. Users often provide exact exon coordinates, and many target reads are rejected because the first aligned chunk falls just outside the defined region. Add at least 500 base pairs of padding on each side. A second common error is using a low complexity genome assembly as the reference. If the reference has gaps or mismapped contigs, the real time aligner will misclassify reads and enrichment will fail. Use the most complete assembly available for the species.

Another mistake is ignoring the pore occupancy rate. During an adaptive sampling run, the sequencer spends time reversing voltage for rejected reads. If you target a very small region, the pore may spend more than 90% of its time rejecting molecules. Your throughput in terms of total base pairs per hour will drop dramatically. For targets smaller than 0.5% of the genome, consider pooling samples with a barcoding approach to increase the absolute number of target reads per run.

Some users also forget to size select their library. Short fragments pass through the pore quickly, and the decision window may end before the basecaller generates enough data for confident alignment. The result is that many short reads are accepted by default but contribute very little to coverage depth. Size select above five kilobases to give the adaptive sampling algorithm enough time to make an informed decision.

Limits Of Interpretation

Adaptive sampling is an enrichment method, not a capture method. You cannot expect to sequence every molecule from your target region with 100% efficiency. The enrichment factor varies between runs and depends on the size and uniqueness of the target. You will still have off target reads in your dataset. The method is also biased against GC rich regions because nanopore sequencing itself has a mild GC bias. If your target region has extreme GC content, consider using a supplementary method such as CRISPR based Cas9 enrichment to boost coverage in that specific area.

You cannot use adaptive sampling to discover novel sequences that are not in the reference genome. Because the decision to accept or reject depends on alignment to a known reference, any molecule that originates from an uncharacterized structural variant or a foreign genome will be rejected. For metagenomic applications where you want to enrich a known pathogen from a host background, adaptive sampling works well Res Sq. But for discovery of completely novel genomes, you should use a standard shotgun approach.

The method also does not preserve absolute quantitative information. Because the rejection rate depends on the sequence content of each molecule, you cannot use read counts as a measure of copy number without extensive normalization. For relative quantification across multiple targets in the same run, you can use normalization against known diploid regions, but absolute copy number calls require careful calibration.

Frequently Asked Questions

How large should my target region be for adaptive sampling to work well?

A target region between 1% and 5% of the genome size typically gives a good balance between enrichment factor and throughput. For the human genome, that means 30 to 150 megabases of target space. Smaller targets down to 0.5% can work but will produce lower total target yield per flow cell.

Can I use adaptive sampling with a reference genome from a closely related species?

Yes but enrichment efficiency will drop. The real time aligner needs high sequence identity to classify a read with confidence. If the reference genome is from a different species, many reads will map ambiguously and the pore may reject target reads. Use the most closely related complete assembly available and add more flanking padding to account for mismatches.

Do I need a special flow cell for adaptive sampling?

No. Adaptive sampling works on standard MinION and PromethION flow cells with R9.4.1 or R10.4.1 chemistry. The adaptive sampling algorithm is implemented in the MinKNOW software. You do not need any custom hardware modifications. However, the flow cell must have a sufficient number of active pores for real time enrichment to be practical.

How does adaptive sampling compare to targeted PCR based nanopore sequencing?

PCR based targeted sequencing can achieve very high coverage depth across small regions at low cost. Adaptive sampling gives you longer reads and avoids PCR bias, but it requires more sequencing time and a larger flow cell investment for small targets. For clinical validation of a single variant in a handful of samples, PCR based amplicon sequencing is often simpler. For discovering structural variants across many loci, adaptive sampling is the better choice Fam Cancer.

References And Further Reading

Galaxy Training Network: Adaptive Sampling Workflow
Tutorial covering the end to end adaptive sampling pipeline using Galaxy.

EMBL EBI Training: Long Read Sequencing Analysis
Course materials for downstream analysis of long read data including base modification calling.

Bioconductor: Long Read Analysis Packages
Open source tools for aligning, visualizing, and calling variants from nanopore data.

Nature Scientific Reports: Imprinting Disorders with Adaptive Sampling
Demonstration of combined genomic and epigenomic profiling using adaptive sampling.

BMC Genomics: DNA Methylation Analysis in Pigs
Study showing adaptive sampling for targeted methylation analysis in a non human genome.

Fam Cancer: SVA Insertion Detection by Long Read Sequencing
Clinical case where adaptive sampling identified a pathogenic retrotransposon insertion.

Research Square: African Swine Fever Virus Genome
Application of adaptive sampling for pathogen genome enrichment from a complex sample.

Integrative and Comparative Biology: Gut Microbiome Dynamics
Metagenomic study using adaptive sampling to enrich specific microbial genomes.

Related Articles

Protein Synthesis
Incomplete Dominance Gene
Cell Membrane Function Biology
Dna Structure
Protein Structure