Deaminase-based RNA recording enables high throughput mutational profiling of protein-RNA interactions

Rachael A. Bakker, Oliver B. Nicholson, Heungwon Park, Yu-Lan Xiao, Weixin Tang, Arvind Rasi Subramaniam, and Christopher P. Lapointe

Preprint	bioRxiv 10.1101/2025.04.11.648485
Sequencing data	NCBI BioProject PRJNA1255650
Per-run accessions	sra_accessions.csv
Plasmid sequences	data/plasmid_maps
Per-panel data	source_data
Questions and reagent requests	open an issue
Licence	MIT for the code, CC BY 4.0 for the data

Figures

Panel	Figure	Code	Source data
1C	reporter_editing	plot_reporter_editing.py	figure_1c.csv
2C	recorder_edit_counts_enzyme_concentration	plot_enzyme_comparisons.py	figure_2c.csv
2D	recorder_site_distance_tada_lambdaN_250nM recorder_site_distance_tada_250nM	plot_recorder_site_distances.py	figure_2d.csv
2E	recorder_context_editing_A8	plot_recorder_contexts.py	figure_2e.csv
3B	boxb_all_variant_A8_vs_A2_250nM	plot_boxb_substitutions.py	figure_3b.csv
3C	boxb_mutation_distance_editing_mean_A8_A2_250nM	plot_boxb_mutation_distance.py	figure_3c.csv
3D	boxb_single_mutant_editing_mean_A8_A2_250nM	plot_boxb_substitutions.py	figure_3d.csv
3E	boxb_lambdan_structure	rendered from the Legault et al. 1998 coordinates, obtained from that paper's corresponding author	—
3F	boxb_multimutant_heatmaps_mean_A8_A2_lambdaN_250nM	plot_boxb_multimutant_heatmaps.py	figure_3f.csv
4B	recorder_edit_counts_pag_vs_nanobody_250nM	plot_enzyme_comparisons.py	figure_4b.csv
4C	recorder_site_editing_pag_components_A8_250nM	plot_enzyme_comparisons.py	figure_4c.csv
4D	recorder_site_editing_pag_conditions_A8	plot_enzyme_comparisons.py	figure_4d.csv
4E	boxb_all_variant_A8_vs_A2_editors	plot_boxb_substitutions.py	figure_4e.csv
4F	boxb_mutation_distance_editing_mean_A8_A2_editors	plot_boxb_mutation_distance.py	figure_4f.csv
4G	boxb_single_mutant_editing_mean_A8_A2_editors	plot_boxb_substitutions.py	figure_4g.csv
5B	boxb_read_coverage_tada_lambdaN_original	plot_boxb_read_coverage.py	figure_5b.csv
5C	recorder_edit_counts_editors_original	plot_recorder_edit_counts.py	figure_5c.csv
5D	recorder_sites_and_placements_tada_lambdaN_original	plot_recorder_site_placement.py	figure_5d.csv
5E	boxb_all_variant_A8_vs_A2_original	plot_boxb_site_correlation.py	figure_5e.csv
5F	boxb_mutation_distance_mean_A8_A2_original	plot_boxb_mutation_distance.py	figure_5f.csv
5G	boxb_single_mutant_heatmap_mean_A8_A2_original	plot_boxb_effect_heatmaps.py	figure_5g.csv
6B	variant_read_coverage	plot_variant_read_coverage.py	figure_6b.csv
6C	recorder_site_profile	plot_recorder_site_profile.py	figure_6c.csv
6D	amino_acid_site_correlation_A2_A15	plot_amino_acid_site_correlation.py	figure_6d.csv
6E	variant_class_distribution	plot_variant_class_distribution.py	figure_6e.csv
6F	position_editing_profile	plot_position_editing_profile.py	figure_6f.csv
6G	boxb_lambdan_structure_dms_positions	as above, with the λN positions in dms_position_summary.csv	—
6H	amino_acid_editing_logo	plot_amino_acid_editing_logo.py	figure_6h.csv
S1A	boxb_hairpin_site_editing_enzyme_concentration	plot_boxb_hairpin_sites.py	figure_s1a.csv
S1B	recorder_edit_counts_tada_lambdaN_time	plot_enzyme_comparisons.py	figure_s1b.csv
S2A	recorder_site_distance_tada_lambdaN_125nM recorder_site_distance_tada_125nM	plot_recorder_site_distances.py	figure_s2a.csv
S2B	recorder_site_distance_tada_lambdaN_500nM recorder_site_distance_tada_500nM	plot_recorder_site_distances.py	figure_s2b.csv
S3A	recorder_context_editing_sites	plot_recorder_contexts.py	figure_s3a.csv
S3B	rabe_rna_sequence_context	plot_context.r	figure_s3b.csv
S4A	boxb_all_variant_A8_vs_A2_125nM	plot_boxb_substitutions.py	figure_s4a.csv
S4B	boxb_mutation_distance_editing_mean_A8_A2_125nM	plot_boxb_mutation_distance.py	figure_s4b.csv
S4C	boxb_single_mutant_editing_mean_A8_A2_125nM	plot_boxb_substitutions.py	figure_s4c.csv
S4D	boxb_all_variant_A8_vs_A2_500nM	plot_boxb_substitutions.py	figure_s4d.csv
S4E	boxb_mutation_distance_editing_mean_A8_A2_500nM	plot_boxb_mutation_distance.py	figure_s4e.csv
S4F	boxb_single_mutant_editing_mean_A8_A2_500nM	plot_boxb_substitutions.py	figure_s4f.csv
S5A	boxb_multimutant_heatmaps_mean_A8_A2_lambdaN_125nM	plot_boxb_multimutant_heatmaps.py	figure_s5a.csv
S5B	boxb_multimutant_heatmaps_mean_A8_A2_lambdaN_500nM	plot_boxb_multimutant_heatmaps.py	figure_s5b.csv
S6A	boxb_single_mutant_editing_mean_A8_A2_time	plot_boxb_substitutions.py	figure_s6a.csv
S6B	boxb_mutation_distance_editing_mean_A8_A2_time	plot_boxb_mutation_distance.py	figure_s6b.csv
S6C	boxb_all_variant_A8_vs_A2_time	plot_boxb_substitutions.py	figure_s6c.csv
S6D	boxb_single_mutant_A8_vs_A2_time	plot_boxb_substitutions.py	figure_s6d.csv
S7B	recorder_site_editing_pag_components_A2_250nM	plot_enzyme_comparisons.py	figure_s7b.csv
S7C	recorder_site_editing_editors_A8_250nM	plot_enzyme_comparisons.py	figure_s7c.csv
S7D	recorder_site_editing_editors_A2_250nM	plot_enzyme_comparisons.py	figure_s7d.csv
S7E	recorder_site_editing_pag_conditions_A2	plot_enzyme_comparisons.py	figure_s7e.csv
S7F	recorder_edit_counts_pag_conditions	plot_enzyme_comparisons.py	figure_s7f.csv
S7G	recorder_edit_counts_pag_components_250nM	plot_enzyme_comparisons.py	figure_s7g.csv
S7H	boxb_direct_vs_split_recruitment_mean_A8_A2	plot_boxb_recruitment_comparison.py	figure_s7h.csv
S8A	recorder_edit_counts_tada_lambdaN_vs_no_editor_original	plot_recorder_edit_counts.py	figure_s8a.csv
S8B	boxb_in_vitro_vs_in_vivo_mean_A8_A2	plot_in_vitro_comparison.py	figure_s8b.csv

Reproduce

Everything runs from the repository root through run_everything.sh, which launches each analysis as a Snakemake workflow:

./run_everything.sh

./run_everything.sh --plots

The first command counts reads and then redraws; the second redraws figures from count tables that are already present. --only <name> restricts either to one analysis, --singularity runs every step in a container, and a trailing rule name runs one figure. The options may be given in any order, and anything after them is passed through to Snakemake.

What has to be installed

Snakemake, and a Python environment holding pandas, numpy, matplotlib, scipy and biopython. By default every step runs with whatever is on PATH, so nothing further is needed on an ordinary machine. The published figures were produced under Snakemake 7.0.0 with Python 3.12.8, pandas 2.3.1, numpy 1.26.4, matplotlib 3.10.3, scipy 1.16.0 and biopython 1.85.

Adding --singularity instead runs each step inside the container its workflow declares, so that the versions above are fixed for you:

./run_everything.sh --singularity

Snakemake then pulls each docker:// image once into the directory named by SINGULARITY_PREFIX and reuses it from then on, so the first run of an analysis spends a few extra minutes building images. Apptainer works in place of Singularity.

Container	Used by
r_python:2.4.2	every workflow, apart from the two below
python:1.0.0	the in vitro read counter
hisat2_3n:1.0.1	the published-data reanalysis behind Supplementary Figure 3B
sratools:3.0.8	download_from_sra.sh, which runs fasterq-dump inside it

Every analysis behind the main figures needs nothing but Python. The published-data reanalysis behind Supplementary Figure 3B is the exception, because it runs HISAT-3N and samtools: give it --singularity, or put those programs on PATH. The launcher checks for them and says which is missing before starting anything.

Both scripts read the same settings from the environment, and each one has a working default:

Variable	Read by	Default	Meaning
SNAKEMAKE	run_everything.sh	snakemake	the Snakemake executable
CORES	run_everything.sh	4	cores the workflows may use
THREADS	download_from_sra.sh	4	threads each download may use
SINGULARITY_PREFIX	both scripts, with --singularity	.singularity	directory that caches the container images
SINGULARITY_ARGS	both scripts, with --singularity	none	arguments passed through to Singularity, for whatever bind mounts the machine needs
SINGULARITY	download_from_sra.sh, with --singularity	singularity	set this to apptainer instead where that is what is installed

On a cluster, containers avoid installing anything: bind the filesystem that holds the data, and point the image cache at shared storage so that the images are built once for everyone.

export SINGULARITY_PREFIX=/path/to/shared/singularity
export SINGULARITY_ARGS="-B /fh"
export CORES=12
./run_everything.sh --singularity

Count tables, to redraw the figures

Count tables for --plots are attached to the count-tables-v1 release, fetched by download_count_tables.sh:

./data/download_count_tables.sh

That is a 67 MB download, and covers every figure except 1C, whose counts are versioned in the repository. Redrawing from the tables takes minutes.

Raw reads, to count from scratch

Raw reads come from the BioProject, fetched by download_from_sra.sh:

./data/download_from_sra.sh

This fetches each run by accession and names the files as the workflows expect. Naming sample identifiers from the sample_id column of sra_accessions.csv fetches only those runs. It calls fasterq-dump from PATH, which conda installs with conda install -c bioconda sra-tools; --singularity runs it inside the container instead, so that nothing has to be installed for it either.

Counting from the reads needs a compute allocation, and far more disk than it downloads:

• The 25 runs are a 6.4 GB download, 20.9 billion bases in all.
• They unpack to 95 GB of FASTQ, because the counting workflows read uncompressed reads. Only the λN DMS mRNA run stays gzipped.
• Counting writes per-barcode intermediates on top of that, leaving 167 GB in the ignored data directory of each analysis folder, 156 GB of it under the in vitro analysis.
• A complete run of the four analyses behind Figures 1C and 2 to 6 took 1 hour 55 minutes on 12 cores, divided about evenly between downloading and counting, and never passed 2 GB of memory.

The published-data reanalysis behind Supplementary Figure 3B is separate, and much heavier: building the HISAT-3N index for GRCh38 alone asks for 300 GB of memory and 16 cores.

Layout

Folder	Contents
nanopore_reporter_editing	Low-throughput reporter editing, Figure 1C
boxb_in_vitro_sequencing	Purified-enzyme boxB library, Figures 2-4 and Supplementary 1-7
boxb_in_vivo_sequencing	Cellular boxB library, Figure 5 and Supplementary 8
ln_dms_sequencing	λN deep mutational scan, Figure 6
published_rabe_contexts	Reanalysis of published rABE RNA sequencing, Supplementary 3B
boxb_structure_views	λN-boxB structure panels, Figures 3E and 6G

Within each analysis folder, annotations holds the versioned CSVs that define samples, read layouts, coordinates and figure selections; scripts holds the workflows and the plotting code; figures holds the output; tables holds summary tables; and data holds reads and count tables, which are not versioned.

Summary table	Contents
figure_1c_statistics.csv	Fold changes and Fisher exact tests behind Figure 1C
wildtype_statistics.csv	λN wild-type dispersion, depth contrast, test calibration and robustness series
rabe_rna_sequence_context.csv	Per-context editing in the published rABE samples