The Guide to DAF-seq

DAF-seq (Deaminase-Assisted single-molecule chromatin Fiber sequencing) is a method that enables single-molecule footprinting at near-nucleotide resolution while synchronously profiling single-molecule chromatin states and DNA sequence. It leverages SsDddA, a nonspecific cytidine deaminase from Simiaoa sunii, to selectively stencil single-molecule protein occupancy on intact nuclei. DAF-seq is compatible with both PacBio HiFi and Oxford Nanopore sequencing platforms, and a single-cell version (scDAF-seq) is also available.

The method is described in:

Swanson, Mao, Mallory, Vollger, Bohaczuk, Oliveira, Lyon, Ranchalis, Parmalee, Cohen, Bennett & Stergachis. Mapping single-cell diploid chromatin fiber architectures using DAF-seq. Nature Biotechnology (2025). DOI: 10.1038/s41587-025-02914-3

Quick start

  • Running the experiment? Start with the DAF-seq Protocol for the complete wet lab workflow.
  • Analyzing data? Jump to the DAF-QC Pipeline for computational setup and quality control.
  • New to DAF-seq? See the Glossary for definitions of key terms.

What's here

Wet Lab

  • DAF-seq Protocol -- Complete protocol for nuclei isolation, on-nuclei cytosine deamination, and DNA extraction
  • Primer Design -- Guidelines for designing primers for targeted DAF-seq
  • Protein Purification -- Purification protocols for SsDddA and DddI

Dry Lab

  • DAF-QC Pipeline -- Snakemake pipeline for quality control and initial processing of DAF-seq data

Reference

  • Glossary -- Definitions of key DAF-seq terms and concepts
  • Cite -- Citation information and BibTeX entry

Contributing

You can help improve this guide! Click the edit button () in the top right of any page to suggest changes.

For questions or contributions, see the relevant GitHub repositories:

DAF-seq Protocol

This protocol covers the complete DAF-seq workflow from nuclei isolation through DNA extraction. For primer design guidelines, see Primer Design. For SsDddA and DddI purification, see Protein Purification. After library preparation and sequencing, use the DAF-QC Pipeline for data processing and quality control.

Reagents & Buffers

Buffer A

Final ConcentrationStock ConcentrationVolume stock soln
RNase- DNase- free H2O100%959 uL
15 mM Tris-Cl, pH 8.01 M Tris-Cl, pH 8.015 uL
15 mM NaCl5 M NaCl3 uL
60 mM KCl3 M KCl20 uL
1 mM EDTA, pH 8.00.5 M EDTA, pH 8.02 uL
0.5 mM EGTA, pH 8.00.5 M EGTA, pH 8.01 uL
0.5 mM Spermidine0.5 M Spermidine1 uL
10 nM ZnCl210 uM ZnCl21 uL

Note: Buffer A without spermidine or ZnCl2 can be prepared in large quantities (e.g. 100 mL) and safely stored for up to 6 months at room temperature. Add 1 uL of 0.5 M spermidine and 1 uL of 10 uM ZnCl2 to 998 uL of buffer A stock right before using. Store spermidine at -20 C for up to 6 months.

Additional reagents

  • Reagents for nuclei isolation, e.g.
    • Buffer A + 0.1% (w/v) digitonin
    • 1x PBS, pH 7.4 + 0.5% BSA
  • UNG inhibitor (NEB M0281)
  • SsDddA stock: 100 uM, aliquots stored at -80 C
  • DddI stock: 1000 uM, aliquots stored at -80 C
  • NEB Monarch Spin gDNA Extraction Kit (T3010)
  • Qubit dsDNA assay
  • LoBind tubes

Nuclei Isolation

Nuclei isolation is cell and tissue-type dependent and should be optimized by the user to ensure completion. Below is an example protocol which we have used for LCLs. Triton and IGEPAL can be substituted for digitonin.

  1. Collect 500k-1M cells into a 1.5 mL LoBind tube.
  2. Pellet at 350 x g for 5 min at 4 C. Remove supernatant.
  3. Wash cells with 1 mL 1x PBS, pH 7.4 + 0.5% BSA.
  4. Pellet at 350 x g for 5 min at 4 C. Remove supernatant.
  5. Resuspend pellet in 60 uL Buffer A.
  6. Add 60 uL 2x Lysis Buffer. Mix by gently tapping the side of the tube and incubate 10 min on ice.
  7. Spin 350 x g for 5 min at 4 C. Remove supernatant.
  8. Resuspend nuclei in 50 uL Buffer A.

IMPORTANT: Ensure that cells are fully lysed before proceeding to the next step.

  1. Count 100k-250k nuclei and add Buffer A + 1 uL UNG inhibitor (2 U) to a final volume of 47 uL.

On-Nuclei Cytosine Deamination

  1. Add 2 uL SsDddA (100 uM stock, final 4 uM) to the 47 uL nuclei suspension.
  2. Incubate 10 min at 25 C. Mix gently; avoid vortexing.
  3. Immediately add 1 uL DddI (1000 uM, 5-molar excess) to stop the reaction.
  4. (Optional) Store at -20 C before DNA extraction.

DNA Extraction

Follow the manufacturer's instructions. The steps below reflect the lab's working sequence for samples after the on-nuclei deamination stop. For routine targeted DAF-seq applications, we recommend NEB Monarch Spin gDNA Extraction Kit (T3010) as the length of isolated DNA is typically sufficient for the amplification of 2-7 kb products. A HMW kit may be necessary for longer range PCR products.

  1. Add PBS to bring the total volume to 100 uL (typically 50 uL).
  2. Add 1 uL Proteinase K and 3 uL RNase A; mix by brief vortexing.
  3. Add 100 uL Cell Lysis Buffer; vortex 10 s.
  4. Incubate 5 min at 56 C in a thermal mixer with agitation (~1400 rpm).
  5. Add 400 uL gDNA Binding Buffer; vortex 5-10 s.
  6. Centrifuge briefly (~1 min at 16,000 x g) if needed to collect contents; transfer to a gDNA purification column seated in a collection tube.
  7. Close cap and centrifuge 3 min at 1,000 x g to load/bind DNA onto the column.
  8. Immediately centrifuge 1 min at 16,000 x g to clear the membrane.
  9. Discard collection tube and place column into a fresh collection tube.
  10. Add 500 uL gDNA Wash Buffer; centrifuge 1 min at >=12,000 x g.
  11. Discard flow-through and repeat the 500 uL Wash Buffer step and spin.
  12. Transfer column to a clean 1.5 mL LoBind tube.
  13. Add 35 uL pre-warmed (56 C) gDNA Elution Buffer directly to the membrane.
  14. Let stand 1 min at room temperature; centrifuge 1 min at >=12,000 x g to elute DNA.
  15. Store eluted gDNA at 4 C (short-term) or -20 C (long-term). Quantify with Qubit.

The DNA is now ready for PCR.

Notes

  • The reaction volume of the "On-Nuclei Cytosine Deamination" may be doubled to accommodate 500K cells. If 100 uL volume is used, no additional buffer is required at step 14 of "DNA Extraction".
  • We have observed that the recovery of SsDddA-treated DNA is approximately 10-20% as efficient as untreated DNA using the Monarch Spin gDNA Extraction Kit. This recovery is typically sufficient for downstream applications.

Next steps

After library preparation and sequencing, process your data with the DAF-QC Pipeline to assess targeting efficiency, deamination rates, strand calling, and other quality metrics.

Primer Design

A strategy for designing primers for targeted DAF-seq experiments.

Materials

  • Untreated gDNA: We recommend a median size of ~20 kb for most applications, which we routinely isolate using the Monarch Spin gDNA Extraction Kit (T3010).
  • DAF-seq DNA: SsDddA-treated chromatin followed by DNA extraction (see DAF-seq Protocol). For primer validation, chromatin from a readily available/cheap source that contains the target locus of interest should be used (e.g. cell lines), although the target sample is preferred when possible.
  • Uracil-tolerant DNA polymerase/PCR master mix: It is critical to use a DNA polymerase that tolerates uracil in the template. We have had the best results with RepliQa HiFi ToughMix (QuantaBio, 95200). Other uracil-tolerant polymerases we have tested with varying results include Q5U (NEB M0597), PhusionU (Thermo Scientific F562), LongAmp (NEB M0323), and KOD One (Toyobo KMM-101).
  • PCR cleanup kit: We use Monarch Spin PCR & DNA Cleanup Kit (NEB T1130), but most kits should work as long as they can process the proper PCR fragment size.

Identify a target window

Targeted DAF-seq is typically performed to amplify 2-10 kb targets using PCR (see Choosing a product length below). We recommend that primer binding sites are designed to be at least 100 bp outside of the genomic element of interest (preferably >500 bp). Below are features to consider when designing primers (in order of priority):

  1. When possible, keep multiple regulatory elements within the same amplicon if co-actuation is to be considered as part of the analysis.
  2. Avoid placing primer binding sites directly within accessible chromatin regulatory elements within your sample of interest, which can typically be determined using existing orthogonal datasets such as ATAC-seq, DNase-seq, or Fiber-seq.
  3. Avoid placing primer binding sites within repetitive elements, which can be identified using the RepeatMasker track in the UCSC genome browser.
  4. Avoid placing primer binding sites directly over common germline genetic variants within your species of interest. If you are using human samples for which a germline VCF is not available, avoid common variants, which can be displayed in the UCSC browser gnomAD track (be sure to select "All" annotation types; we recommend a threshold of 0.01).
  5. Consider avoiding primer binding sites that result in internal repetitive elements with multiple copies within the amplified region. In some cases, homologous repeats can cause deletions within PCR products. This is easily detected by shallow sequencing and should not be prioritized in the first round of design if it compromises inclusion of desired elements.

Design candidate primers

Primer design for DAF-seq requires unique considerations due to template deamination. The guidelines below represent a primer design strategy using NCBI primer-BLAST followed by curation.

As each region and deamination pattern is unique, it may not always be possible to satisfy all recommendations:

  1. Set design parameters: Paste the target region and specify the positions in which to design primers. Set the PCR product size to the desired range. You may also broaden the "Primer melting temperatures" if few primers are returned.

    In the "Primer Pair Specificity Checking Parameters", change "Database" to "Refseq representative genomes", and increase the "Max target amplicon size" as appropriate (typically a few kb higher than the intended target).

    In "Advanced parameters", change "Primer Size" to 20-28 (optimal 23). Change "Primer GC content(%)" to 20-50%. Change "Max Poly-X" to 3 (see Poly-X stretches). Change "Max GC in primer 3' end" to 3.

  2. Curate primers for the following:

    1. Specificity (critical): For primers with identified off-target sites, pay special attention to those where a C>T mutation would make a better primer binding site. Reject primer candidates with many partial off-target matches.
    2. C/G in 3' end (critical): Prefer primers with the absence of C/G in the terminal 3 bases on the 3' end, although primers with the absence of C/G in the terminal 2 bases have also been successful for DAF-seq.
    3. Minimize C/G content (preferred): When possible, prefer primers with <40% C/G.
    4. Balance C/G (optional): See C and G in primers.

  3. Select primer sets for testing: From our experience, less than 50% of selected primer sets will successfully amplify a single band with SsDddA-treated template. For this reason, we recommend selecting several primer sets per target to accelerate screening.

PCR annealing temperature gradient with gDNA

Critical: Ensure that the DNA polymerase tolerates uracil in the template. We have had the best results with RepliQa HiFi ToughMix (QuantaBio, 95200). Other uracil-tolerant polymerases include Q5U, PhusionU, LongAmp, and KOD One.

Perform an annealing temperature gradient on primer candidates using untreated gDNA template. We prefer to use gDNA for validation at this point both because gDNA is cheaper and easier to produce and because primer binding sites are intentionally designed outside of known regulatory elements, so the majority of the final DAF-seq template should be nucleosome-protected and unmodified, as is simulated by gDNA. Basically, if it doesn't work on gDNA, it won't work for DAF-seq.

Look for primer sets that produce a single band of the anticipated size in agarose gel electrophoresis. Low off-target amplification may be tolerable if primer choices are limited. If multiple temperatures produce comparable amplification of the single product, we prefer the lowest temperature which should theoretically have more tolerance for deaminations in DAF-seq DNA.

If multiple primer sets are successful at this stage, we recommend comparing the performance of all of them through sequencing on DAF-seq template.

Validate PCR with DAF-seq template and shallow sequencing

Perform PCR amplification on DAF-seq template with the best primer sets and annealing temperatures determined in the previous step. We recommend a 50 uL reaction. Run ~10 uL on an agarose gel to validate the presence of a single band. Off-target amplification may be tolerable if primer choices are limited.

Purify the remaining product with a PCR cleanup kit. We do not recommend gel extraction as the buffers can interfere with subsequent sequencing.

Perform shallow sequencing with a long-read method. We typically use the ONT-based Plasmidsaurus Premium PCR (standard) for routine validation. In our experience, primers that perform well in shallow ONT-based sequencing have also performed well on full-scale PacBio sequencing.

Run QC to evaluate primer performance

Run the DAF-QC-SMK pipeline to evaluate primer performance. Optimal primers will have:

  • Low (<10%) off-target amplification
  • A high proportion of full-length reads (>80%)
  • Approximately 50/50 ratio of CT/GA reads, although primer sets with substantial strand bias may be acceptable for some applications (such as when identifying single nucleotide variants is not critical)

Also be sure to visualize reads in IGV. Check for any internal deletions and evaluate whether they are haplotype-specific. Non-haplotype-specific deletions may represent PCR artifacts that can be resolved by tuning PCR conditions or redesigning primer binding sites to exclude problematic repetitive regions (see point 5 in Identify a target window).

Additional considerations

Primer design rules are not rigid

Primer design for DAF-seq can be much more difficult than typical PCR primer design. Identifying primer sets with high specificity is challenged by the presence of template modifications and the preference to minimize C/G content. For the most difficult targets, a flexible approach that bypasses some of the recommendations above may be necessary.

Nucleosome protection is an asset

When a regulatory landscape is known through alternative datasets, it is important to avoid primer binding sites overlapping accessible elements. With this in mind, most of the template DNA will be nucleosome-occluded as nucleosome footprints are ~147 bp, while internucleosomal linker regions (i.e. accessible regions between nucleosomes that are not regulatory elements) are closer to ~10 bp. This means that most template strands will be complementary to primers even within C/G positions.

Degenerate bases in primers

To account for the possibility that there could be deaminations within primer binding sites, our initial publication used primers designed with degeneracy at G genomic positions in the form of randomly incorporated G or A (IDT). Given the point above regarding nucleosome protection, we later reconsidered this design and tested whether degenerate bases facilitate amplification. In general, we found that targets amplified better without degenerate bases.

Selection of PCR polymerase

It is critical to select a polymerase tolerant of uracil in the DNA template. This is one of the most common mistakes. Polymerase fidelity will also impact the quality of results, especially if deduplication is desired as spurious mutations can cause duplicates to be missed. The polymerase must also be suitable for relatively long-range amplifications, although this is quite common. We have found that RepliQa HiFi ToughMix (QuantaBio, 95200) fulfills these requirements and has been relatively robust in our hands.

Locked nucleic acids (LNAs) for difficult targets

In specialized cases, we have used primer sets containing locked nucleic acids to boost product yield and annealing temperature (e.g. when needed for compatibility in multiplexed assays or for C/G-rich templates prone to forming secondary structures). We have found that primers containing a single LNA are tolerated in PacBio and ONT sequencing, but adding more than one LNA causes substantial sequencing dropout with PacBio (ONT not tested). As LNA-containing primers are substantially more expensive, we recommend adding them only when a faint on-target band is already observed and multiple primer sets have failed.

In our hands, we have observed improved yield with LNAs positioned within 5 bases of the 3' and 5' ends, and we typically add them between A/A, A/T, or T/T bases. The effects of LNAs on primer specificity are complex, and LNA-containing primers have only been superficially explored in our lab. We recommend further literature review for users requiring LNA-containing primers.

Multiplexed assays

We have successfully designed primer sets for multiplexed PCR, although this can be extremely challenging. Try to design primers with comparable predicted annealing temperatures and then test empirically using a PCR annealing temperature gradient with gDNA template. LNAs can be used to boost primer annealing temperatures if needed.

Once compatible annealing temperatures have been established, try the multiplex PCR using an equimolar ratio of primers, purify the product, and sequence with shallow long-read sequencing (i.e. Plasmidsaurus Premium PCR sequencing or similar). Check that the sequencing yield of each target is as desired (typically equal, but you may wish to prioritize a specific amplicon for some applications). Tune primer ratios by titration if needed, and reconfirm with sequencing.

Choosing a product length

Target amplicon length is typically 2-10 kb. The lower limit is set to include multiple nucleosome-bound stretches which are important for genomic alignment, and we do not recommend smaller for most applications. The upper limit is more flexible, but constrained by the same challenges as conventional long-range PCR. However, primer specificity can be even more critical because deamination events within the template can create stronger off-target binding sites.

Poly-X stretches

Details on poly-X stretch considerations are being finalized.

C and G in primers

A matter of ongoing discussion is whether to preferentially avoid G bases only or both G and C bases in primer design. SsDddA converts cytosine to uracil bases, so G bases in the primers have the potential for mismatch upon the first PCR cycle. However, on the second cycle, SsDddA-converted uracil bases that were on the 3' end of the PCR-synthesized strand are now A bases at positions complementary to C within the primer. Some lab members have prioritized the first round of amplification by preferentially excluding G. Others argue that an equal C/G ratio in primers is important to maximize efficient amplification on both the first and second cycles, which is important for exponential amplification and anecdotally reduces biases for top vs bottom strands. In practice, the exact choice may not be critical as the majority of DAF-seq template is anticipated to be nucleosome-protected and unmodified at primer binding locations, and DAF-seq primers have been successfully designed with both strategies.

Feedback

DAF-seq is an emergent technology, and we are still improving our primer design methodology. If you have suggestions or collaboration ideas for making this process more robust, please reach out.

Protein Purification

This section is under development. Protocols for purifying SsDddA and DddI will be added soon.

Planned content includes:

  • Expression and purification of SsDddA (the Simiaoa sunii cytidine deaminase used in the DAF-seq Protocol)
  • Expression and purification of DddI (the deaminase inhibitor used to stop the reaction)
  • Recommended storage conditions and aliquoting strategies
  • Quality control assays for verifying enzyme activity

In the meantime, refer to the methods section of the DAF-seq publication for purification details.

DAF-QC Pipeline

DAF-QC-SMK is a Snakemake pipeline for quality control and initial processing of DAF-seq sequencing reads. It supports both PacBio HiFi and Oxford Nanopore platforms. This page covers installation, usage, and key outputs. For the wet lab steps that precede this pipeline, see the DAF-seq Protocol.

Getting started

The pipeline uses pixi for environment management. Clone the repository and install:

git clone https://github.com/StergachisLab/DAF-QC-SMK.git
cd DAF-QC-SMK
pixi install

Verify the installation

A test dataset (human chr8, hg38) is bundled with the repository. Run it to confirm everything is working before processing your own data:

pixi run test

If you encounter errors, please run the test case before contacting the developers, as it helps with troubleshooting.

Usage

Run the pipeline with pixi:

pixi run snakemake --configfile config/config.yaml

For SLURM clusters, specify a profile:

pixi run snakemake --configfile config/config.yaml --profile profiles/slurm-executor

You can also run the pipeline from a different directory using --manifest-path:

pixi run --manifest-path /path/to/DAF-QC-SMK/pixi.toml snakemake --configfile config/config.yaml

Inputs

The pipeline requires two configuration files:

Sample table (config.tbl)

A tab-separated table with sample name, BAM/FASTQ path, and targeted regions:

sample	file	regs
test    test.bam    chr8:144415767-144417958

For PacBio BAM inputs, files should contain either unaligned reads or primary reads only (for compatibility with pbmarkdup during consensus generation). See config/config.tbl in the repository for a template.

Configuration file (config.yaml)

Specifies paths to the sample table and reference genome, sequencing platform, and optional parameters:

ref: /path/to/genome.fa
manifest: config/config.tbl
platform: pacbio  # 'pacbio' or 'ont'

# Optional (both platforms)
chimera_cutoff: 0.9
min_deamination_count: 50
end_tolerance: 30
decorated_samplesize: 5000

# PacBio-specific
consensus: True
consensus_min_reads: 3

# ONT-specific
is_fastq: False

See config/config.yaml in the repository for the full list of options with descriptions.

Key outputs

  • Aligned BAMs: Primary, supplementary, and unaligned reads with PCR duplicates marked (du and ds tags).
  • Decorated BAMs: Full-length reads with top/bottom strand designation (C-to-T as top strand, G-to-A as bottom strand). Strand stored in the st tag.
  • Consensus BAMs (PacBio only): MSA consensus of full-length, strand-designated reads. The dc tag indicates the number of reads used to construct each consensus.
  • QC metrics: Targeting efficiency, deamination rates (overall and by 2-bp sequence context), strand calling, enzyme bias, and mutation rates.
  • HTML dashboard: results/{sample_name}/qc/{sample_name}.dashboard.html with all QC plots. The dashboard is self-contained (plots are embedded), so you can copy a single file for sharing or local viewing.

Downstream analysis with fibertools

After QC, DAF-seq data can be further processed with fibertools (ft) for chromatin fiber analysis:

  1. ft ddda-to-m6a: Converts DAF-seq deamination marks (C-to-T / G-to-A) into m6A-equivalent format, enabling compatibility with the Fiber-seq analysis ecosystem.
  2. ft add-nucleosomes: Infers nucleosome positions from the converted deamination data.

These steps allow you to use the full suite of Fiber-seq visualization and analysis tools on DAF-seq data. See the fibertools documentation for details.

Further reading

See the DAF-QC-SMK README for full details on all configuration options, output file formats, and BAM tag specifications.

Glossary

DAF-seq (Deaminase-Assisted single-molecule chromatin Fiber sequencing): A method for single-molecule footprinting at near-nucleotide resolution that simultaneously profiles chromatin states and DNA sequence. It uses cytosine deamination on intact nuclei to stencil protein occupancy along individual DNA fibers.

SsDddA: A nonspecific double-stranded DNA cytidine deaminase from Simiaoa sunii. In DAF-seq, SsDddA is applied to intact nuclei where it deaminates accessible (non-protein-bound) cytosines to uracils, which appear as C-to-T transitions after sequencing.

DddI: A protein inhibitor of DddA-family deaminases. In DAF-seq, DddI is added at a 5-fold molar excess over SsDddA to stop the deamination reaction.

UNG inhibitor: Uracil-N-glycosylase inhibitor (UGI). Added before the deamination step to prevent endogenous UNG from removing uracils created by SsDddA, which would cause DNA strand breaks and data loss.

scDAF-seq: Single-cell DAF-seq. A variant of DAF-seq that uses primary template-directed amplification (PTA) to enable single-cell chromatin fiber profiling.

Decorated BAM: A BAM file in which full-length DAF-seq reads have been assigned a top or bottom strand designation based on deamination patterns (C-to-T for top strand, G-to-A for bottom strand). Strand designation is stored in the st BAM tag.

Chimera: A sequencing read that contains deamination signal from both strands, indicated by a mixture of C-to-T and G-to-A conversions. Chimeric reads are filtered using a configurable cutoff (default: 90% single-strand signal).

Full-length read: A read that spans the entire targeted amplicon (within a configurable end tolerance, default 30 bp). Only full-length reads are used for strand designation and downstream analysis.

Consensus BAM (PacBio only): A BAM file containing multiple-sequence-alignment (MSA) consensus sequences built from groups of PCR duplicate reads. Consensus generation improves accuracy of deamination calls.

fibertools: A suite of command-line tools for analyzing single-molecule chromatin fiber data. The ddda-to-m6a subcommand converts DAF-seq deamination marks (i.e. Y and R bases) into m6A-equivalent format for compatibility with Fiber-seq analysis tools, and add-nucleosomes infers nucleosome positions from the resulting data.

Cite

If you use DAF-seq in your research, please cite the following:

Swanson, E. G., Mao, Y., Mallory, B. J., Vollger, M. R., Bohaczuk, S. C., Oliveira, C. B., Lyon, D. B., Ranchalis, J., Parmalee, N. L., Cohen, B. A., Bennett, J. T. & Stergachis, A. B. Mapping single-cell diploid chromatin fiber architectures using DAF-seq. Nature Biotechnology (2025). https://doi.org/10.1038/s41587-025-02914-3

BibTeX

@article{swanson2025dafseq,
  title   = {Mapping single-cell diploid chromatin fiber architectures using {DAF}-seq},
  author  = {Swanson, Elliott G. and Mao, Yizi and Mallory, Benjamin J. and
             Vollger, Mitchell R. and Bohaczuk, Stephanie C. and
             Oliveira, Christopher B. and Lyon, Daniel B. and Ranchalis, Jane and
             Parmalee, Nancy L. and Cohen, Barak A. and Bennett, James T. and
             Stergachis, Andrew B.},
  journal = {Nature Biotechnology},
  year    = {2025},
  doi     = {10.1038/s41587-025-02914-3},
  url     = {https://doi.org/10.1038/s41587-025-02914-3}
}