Allele-specific analysis from RNA-seq is a powerful approach to characterize cis-regulatory effects. However, existing methods remain limited in both haplotype inference and allelic testing. Their haplotype-inference workflows separate variant calling, haplotype phasing, and read-haplotype assignment into sequential steps, failing to fully exploit within-read single-nucleotide variant (SNV) linkage information and propagating errors into downstream allelic analysis. At the testing stage, they ignore non-phasable reads lacking heterozygous SNVs, biasing calls and inflating false positives, and remain incomplete across gene-, isoform-, and local-event-level variant effects.

Here, we present LongAllele, a statistical framework that employs an expectation–maximization algorithm to jointly infer heterozygous variants, haplotype structure, and read-haplotype assignments from long-read bulk and single-cell RNA sequencing. LongAllele further introduces phasability-aware testing that explicitly accounts for non-phasable reads, avoiding inflated false-positive calls when haplotype information is incomplete. It also enables comprehensive allelic testing across gene-level allele-specific expression (ASE), isoform-level allele-specific transcript usage (ASTU), and local-event-level haplotype-associated exon and junction usage (HAEU and HAJU), providing a multi-scale view of cis-regulation. LongAllele offers a unified framework for haplotype-resolved cis-regulatory analysis across diverse cellular contexts.

Installation

git clone https://github.com/WGLab/LongAllele.git
cd LongAllele
# Recommended: a fresh Python 3.9+ environment (conda or venv)
pip install -r requirements.txt

Upstream dependency: SCOTCH is required to produce the scotch_target directory consumed by LongAllele.

Quickstart

LongAllele runs on HPC clusters via SLURM. The bundled longallele.sh script submits all pipeline steps as a dependency-chained job graph — one command is all you need.

1. Fill in your paths and basic parameter settings:

cp config_template.sh my_run.sh
# edit my_run.sh — set SCOTCH_TARGET, BAM_PATH, REF_FASTA, OUTPUT_DIR, etc.

2. Submit the full pipeline:

bash longallele.sh my_run.sh

That's it. All steps are submitted with correct SLURM dependencies and run automatically in order. Output lands in OUTPUT_DIR once all jobs complete.

See Per-step pipeline for per-step documentation and all configurable arguments.

Per-step pipeline

LongAllele consists of five sequential steps. The pipeline takes an aligned BAM and reference FASTA (with SCOTCH read-to-isoform mappings as upstream input) and produces per-gene haplotype statistics, haplotype-aware count matrices (gene- and isoform-levels), and downstream allelic effect-size and SNV–event linkage tables. Each step section below contains a ▶ Configurable arguments panel — click to expand for the full parameter list. Steps can also be run individually with python src/longallele.py --task <step> without using longallele.sh.

All genomic positions in LongAllele outputs use 0-based coordinates.

BAM + FASTA
    ↓
SCOTCH  ───────────────────────────────────────────────┐
    ↓ (read→gene/isoform mappings)                     │
Step 1: Variant calling (initial SNV candidates)       │
    ↓                                                  │
Step 2: EM input generation (per-gene read×SNV tables) │
    ↓                                                  │
Step 3: EM haplotyping (read→haplotype assignments)    │
    ↓                                                  │
Step 4: Summary statistics + count matrices            │
    ↓                                                  │
Step 5: Downstream analysis ◄──────────────────────────┘
        (effect sizes, SNV–event linkage)

Step 1 — Variant calling

This step generates initial heterozygous SNV candidates from per-gene pileup of the aligned BAM.

Parallelization

SLURM array parallelized across genes. Add --n_jobs N --job_index $SLURM_ARRAY_TASK_ID to fan out across array tasks.

Configurable arguments

Required

Parameter	Description
--task	step1
--scotch_target	Path(s) to SCOTCH output directory (space-separated for multi-sample)
--bam_path	Aligned BAM file(s) (space-separated for multi-sample)
--ref_fasta_path	Reference genome FASTA
--output_folder	Output directory

Optional

Parameter	Description	Default
--depth	Minimum read depth at SNV position	20
--n_alt_count	Minimum alt-allele read count	10
--min_mapq	Minimum mapping quality	20
--min_baseq	Minimum base quality	5
--min_dist_to_end	Minimum distance from read end	3
--prefix	Output filename prefix	none
--gene_subset_path	Restrict to subset of genes (one ID per line)	none
--sample_names / --sample_name_parse	Multi-sample naming overrides	auto
--ref_pickle_path	Pre-built reference pickle	auto
--n_jobs / --job_index	Array parallelization	1 / 0

Job completion

A step1_job{N}.done marker is written to {output_folder}/job_markers/ for each successful array task. After the array finishes, ls {output_folder}/job_markers/step1_*.done | wc -l should equal --n_jobs.

Step 1.5 — Read-block collection (optional)

This step enables the obs_* raw-read validation columns in event_snv.csv (step 5 output). It is not required for core ASE/ASTU analysis — skip it if you do not need raw-read event validation.

Run as a two-part task: a per-sample array followed by a single merge job.

# Part 1 — per-sample array (one task per BAM; run after step 1)
python src/longallele.py --task step1_5 \
    --scotch_target /path/to/scotch_output \
    --bam_path /path/to/aligned.bam \
    --output_folder /path/to/results \
    --n_jobs N_SAMPLES --job_index $SLURM_ARRAY_TASK_ID

# Part 2 — merge (single job; run after part 1 array completes)
python src/longallele.py --task step1_5_merge \
    --scotch_target /path/to/scotch_output \
    --output_folder /path/to/results

longallele.sh handles both parts automatically (run in parallel with step 2, output ready before step 5).

---

Step 2 — EM input generation

This step prepares the per-gene read profile and error profile used as input by the EM in step 3.

Parallelization

SLURM array parallelized across genes (use the same --n_jobs N size as step 1). Add --n_jobs N --job_index $SLURM_ARRAY_TASK_ID.

Configurable arguments

Required — same as Step 1 (--task step2, --scotch_target, --bam_path, --ref_fasta_path, --output_folder).

Optional — same set as Step 1.

Job completion

A step2_job{N}.done marker is written to {output_folder}/job_markers/ for each successful array task. After the array finishes, ls {output_folder}/job_markers/step2_*.done | wc -l should equal --n_jobs.

Step 3 — EM haplotyping

This step jointly infers heterozygous variants, haplotype structure, and read–haplotype assignment per gene via expectation–maximization.

Key flags to set for real data:

• --clf_init — use SNV classifier scores to initialize haplotype priors. Recommended for all real-data runs.
• --high_artifact_mode — enable extra filters for nascent-RNA leak in single-nucleus RNA-seq. Recommended for brain (snRNA-seq) data.

Parallelization

SLURM array parallelized across genes (same --n_jobs N as steps 1–2). Add --n_jobs N --job_index $SLURM_ARRAY_TASK_ID.

Configurable arguments

Required

Parameter	Description
--task	step3
--scotch_target	Path(s) to SCOTCH output directory
--bam_path	Aligned BAM file(s)
--ref_fasta_path	Reference genome FASTA
--output_folder	Output directory

Common optional

Parameter	Description	Default
--prefix	Output filename prefix	none
--cell_type_df_path	CSV with Cell / CellType columns for per-cell-type analysis (example). Cell values must match the cell barcodes in SCOTCH's all_read_isoform_exon_mapping.tsv.	none
--gene_subset_path	Restrict to subset of genes	none
--rna_editing_db	A-to-I editing DB (.npz); override only for non-hg38	bundled hg38
--snv_confidence_path	Pre-defined high-confidence SNV set; skips noise filters	none
--n_jobs / --job_index	Array parallelization	1 / 0
--mtx / --csv	Per-cell count matrix output format	csv

EM tuning

Parameter	Description	Default
--seed	Random seed for EM initialization	42
--max_iter	Maximum EM iterations per gene	50
--tol	Convergence tolerance (parameter delta)	1e-3
--heterozygous_filter	Heterozygosity probability threshold	0.99
--het_fallback	Stepped het-threshold descent	off

When --het_fallback is enabled, the threshold decreases by 0.05 iteratively down to 0.5; the first step yielding any passing SNVs is used, capped at ceil(6.6 × exon_length / 1000) per gene. Intended for simulated / low-coverage data; off by default for real high-coverage data.

SNV noise filtering (pre-EM)

Skipped entirely when --snv_confidence_path is provided.

Filter	What it removes	Parameter	Default
Heterozygous probability	Homozygous sites (binomial model)	--heterozygous_filter	0.99
Low-complexity repeat	Homopolymer / dinuc / trinuc repeats	--repeat_filter_kmer	1
Long alt stretch	Long repeat artifacts	--alt_stretch_filter	50
Variant cluster	Dense SNV runs (escape: alt_count ≥ --alt_cluster_filter is kept)	--var_cluster_window / --var_cluster_n / --alt_cluster_filter	20 bp / 3 SNVs / 150
RNA editing	Known A-to-I editing sites (REDIportal v3)	--rna_editing_db	bundled hg38

--repeat_filter_kmer controls which repeat sizes are checked:

Value	Behavior	Recommended for
0	Disabled	Pre-called confident SNV input
1 (default)	Homopolymer (≥5 consecutive single base)	General use
2	+ Dinucleotide repeats (≥3 copies, e.g. ACACAC)	Stricter filtering
3	+ Trinucleotide repeats (≥3 copies, e.g. AAGAAGAAG)	Most aggressive

⚠️ Trinucleotide filtering can remove genuine heterozygous variants in synonymous codon repeats with functional splicing effects.

The bundled RNA editing database is derived from REDIportal v3 (15.7 M sites, hg38, 24 MB, 0-based, per-chromosome sorted arrays for np.searchsorted lookup). Override --rna_editing_db only for non-hg38 references.

Advanced — SNV classifier (optional)

Pre-EM classifier-based filtering. Train a classifier on validated SNV calls and apply scores to filter or initialize haplotype priors.

Parameter	Description	Default
--snv_classifier	Path to serialized classifier (.joblib)	none
--clf_hard_threshold	Drop SNVs below this classifier score	0.05
--clf_init	Use classifier scores to initialize h_m in EM. Strongly recommended to set for real data.	off
--gap_tau	Gap threshold for adaptive_keep_mask (1.0 = disabled)	1.0
--clf_pruning_threshold	Score below which SNVs are considered low-quality	0.1
--clf_pruning_frac	Max fraction of low-quality SNVs allowed (1.0 = no pruning)	1.0

Advanced — High-artifact mode (snRNA-seq nascent-RNA leak)

Single-nucleus RNA-seq libraries can contain substantial unspliced pre-mRNA (nascent-RNA leak). In long-read snRNA-seq, intron-dominated reads have ambiguous isoform origin and introduce artifacts into haplotype phasing. High-artifact mode (--high_artifact_mode) is an opt-in opt that adds two disabled-by-default filters to mitigate this. Default OFF; output is byte-identical to the standard pipeline when disabled.

• SNV-level filter. LongAllele uses both exonic and intronic SNVs as phasing markers by default. In high-artifact mode, SNVs in read-dense intronic regions of high-leak genes are excluded (controlled by --novel_exon_pct_max), preventing nascent intronic allelic biases from distorting read–haplotype assignment for mature transcripts.
• Read-level filter. Reads whose alignments lie predominantly within introns are also excluded (controlled by --read_intronic_pct_max).

Parameter	Description	Default
--high_artifact_mode	Enable the SNV-level + read-level filters above	off
--novel_exon_pct_max	SNV-level filter cutoff: per-gene fraction of intronic territory broadly covered by reads; genes above the cutoff drop intronic SNVs	0.25
--read_intronic_pct_max	Read-level filter cutoff: per-read intronic / total aligned bp; reads above the cutoff are excluded	0.60
--gsi_base_pkl_path	Explicit path to SCOTCH base pickle (auto-resolved if omitted)	auto

Job completion

A step3_job{N}.done marker is written to {output_folder}/job_markers/ for each successful array task. After the array finishes, ls {output_folder}/job_markers/step3_*.done | wc -l should equal --n_jobs.

Step 4 — Summary statistics + count matrix

This step aggregates per-gene haplotype statistics and produces haplotype-aware count matrices at both gene and isoform levels (bulk + per cell type).

Parallelization

Single job by default, or SLURM array across samples for multi-sample inputs (--job_array_by_sample --job_index $SLURM_ARRAY_TASK_ID).

Outputs

File	Path
summary_statistics.csv	{output_folder}/summary_statistics_{prefix}/summary_statistics.csv
snv_hap_map.csv	{output_folder}/snv_hap_{prefix}/snv_hap_map.csv
read_hap_map.csv	{output_folder}/snv_hap_{prefix}/read_hap_map.csv
Per-gene summaries	{output_folder}/summary_statistics_{prefix}/all_genes_separate/
Bulk isoform count matrices	{output_folder}/count_hap_{prefix}/all_genes/
Per-cell-type isoform tables	{output_folder}/count_hap_{prefix}/ct_isoform_separate/{cell_type}/
Per-cell-type aggregated isoform tables	{output_folder}/count_hap_{prefix}/all_genes/ct_{cell_type}_isoform_agg*.csv

Column dictionary

summary_statistics.csv — per-gene haplotype + isoform statistics

Column	Description
geneID, geneName	Gene identifiers
gamma	EM mapping-bias parameter (fixed at 0.5 in current model)
n_reads, n_reads_phasable, n_snvs	Per-gene read and SNV counts (n_reads_phasable = reads that cover ≥1 phasing SNV)
alpha_hat, alpha_hat_low, alpha_hat_high	Minor haplotype allelic balance (EM point estimate + bounds)
major_hap	Major haplotype label (A or B)
ll_alt, ll_null	Log-likelihoods of the alternative (αEM) and null (α = 0.5) models
lrt_stat, p_value	Likelihood ratio statistic and unadjusted gene-level ASE p-value
p_value_gene_adj	FDR-adjusted gene-level ASE p-value (within sample)
chi2_isoform, df_isoform	Chi-squared statistic and degrees of freedom for the ASTU test
p_value_isoform, p_value_isoform_high, p_value_isoform_low	ASTU p-values (point + bound variants)
p_value_isoform_adj, p_value_isoform_adj_high, p_value_isoform_adj_low	FDR-adjusted ASTU p-values
CellType	Cell type identifier (Bulk for bulk-level rows)

snv_hap_map.csv — per-SNV haplotype assignments

Column	Description
chrom, pos	SNV genomic coordinates (0-based)
ref, alt	Reference and alternate alleles
depth, alt_count, alt_frac	Read support at the SNV site
ID	SNV identifier (chr:pos:ref:alt)
het_prob	Heterozygous probability under the binomial model
h_A	Probability that the alt allele is on haplotype A
h_m	Marker probability — confidence that the SNV is a true heterozygous phasing marker
hat_Z_binary	Binary indicator (1 = SNV retained as a phasing marker after EM)
geneName, geneID	Gene identifiers

read_hap_map.csv — per-read haplotype posteriors

Column	Description
Read	Read name from the source BAM
hat_I	Posterior probability that the read is on haplotype A
hat_I_B	Posterior probability that the read is on haplotype B (= 1 − hat_I)
geneName, geneID	Gene identifiers

Configurable arguments

Required

Parameter	Description
--task	step4
--scotch_target	Path(s) to SCOTCH output directory
--output_folder	Output directory
--summary_haplotype	Write summary_statistics.csv (required for step 5 input)
--summary_count	Write count matrices (bulk + per-cell-type)

Both --summary_haplotype and --summary_count are flags that default to off — without them, step 4 runs but produces no output.

Optional

Parameter	Description	Default
--prefix	Output filename prefix	none
--cell_type_df_path	CSV with Cell / CellType columns	none
--job_array_by_sample	Process one sample per job_index (multi-sample)	off
--job_index	Sample index when --job_array_by_sample is on	0

Job completion

{output_folder}/job_markers/step4.done is written when step 4 finishes (or step4_sample{i}.done per sample when --job_array_by_sample is on).

Step 5 — Downstream analysis

This step computes per-SNV and per-event allelic effect sizes (ASE, ASTU) and haplotype–event association tests, and links nearby SNVs to their events.

Parallelization

Single job, multi-CPU by default (set --n_workers to your available CPU count). For multi-sample inputs, run as a SLURM array across samples (--job_array_by_sample --job_index $SLURM_ARRAY_TASK_ID).

Outputs

File	Description	Path
gene_snv.csv	SNV-centric — phased SNV assignments and signed ASE / ASTU effects, with gene-level allelic statistics per cell type.	{output_folder}/downstream_{prefix}/gene_snv.csv
event_snv.csv	Event-centric — haplotype-associated exon and junction events, linked nearby SNVs, and raw chi-squared validation.	{output_folder}/downstream_{prefix}/event_snv.csv

Column dictionary

gene_snv.csv — SNV-centric table

Column	Description
Sample, CellType	Sample and cell type identifiers
geneID, geneName, geneChr	Gene identifiers
n_reads, n_reads_phasable, gene_n_snvs, gene_n_snvs_called	Gene-level read counts (n_reads, n_reads_phasable) and SNV counts — gene_n_snvs = total candidate SNVs, gene_n_snvs_called = SNVs actually called / used in phasing.
gene_alpha_hat, gene_alpha_hat_low, gene_alpha_hat_high	Minor haplotype allelic balance (EM estimate + bounds)
gene_alpha_hat_major, gene_alpha_hat_major_low, gene_alpha_hat_major_high	Major haplotype allelic balance (1 − minor)
gene_major_hap, gene_minor_hap	Haplotype labels (A or B)
gene_p_value, gene_p_value_adj	Gene-level ASE significance test on phased reads (BH-adjusted). Raw test significance, NOT the final call — small p alone over-calls ASE; final call is ASE_call.
ASE_call	Final ASE call (3-category): 1 significant (gene_p_value_adj ≤ 0.05 and gene_alpha_hat_high < 0.5), -1 not significant (gene_p_value_adj > 0.05), 0 inconclusive (p significant but α CI overlaps 0.5).
dominant_isoform_overall	Most expressed isoform across both haplotypes
top_isoform_hap_major, top_isoform_hap_minor	Top isoform on each haplotype
top_isoform_hap_major_frac, top_isoform_hap_minor_frac	Fraction of hap reads from top isoform
isoform_p_value, isoform_p_value_adj	Gene-level ASTU significance test on phased reads (point estimate, BH-adjusted). Raw test significance, NOT the final call — final call is ASTU_call.
isoform_p_value_high, isoform_p_value_low, isoform_p_value_adj_high, isoform_p_value_adj_low	ASTU significance at the high / low bounds of the isoform-balance CI (BH-adjusted); used to derive ASTU_call.
ASTU_call	Final ASTU call (3-category): 1 significant (isoform_p_value_adj_high ≤ 0.05), -1 not significant (isoform_p_value_adj_low > 0.05), 0 inconclusive (low bound significant but high bound not).
shrinkage_k	Shrinkage constant added to major / minor hap read counts when computing es_ase / es_astu (effect-size regularization; does not shape the CI).
es_ase	ASE effect size: log2(major / minor hap reads)
es_astu	ASTU effect size: log2(dominant isoform major / minor fraction)
astu_source	bulk, ct_specific, or bulk_fallback
snvID	Stable SNV key (chr:pos:ref:alt)
snv_pos, snv_ref, snv_alt	SNV coordinates and alleles
snv_depth_bulk, snv_alt_count_bulk, snv_alt_frac_bulk	SNV read support from variant calling (bulk pileup)
h_A, hat_Z_prob_revised	Haplotype-A frequency and phasing confidence
snv_hap	Haplotype carrying the alt allele (A or B)
snv_on_minor_hap	Whether SNV alt allele is on the minor haplotype
snv_expr_direction	higher_gene_expression or lower_gene_expression
snv_es_ase_signed	Signed ASE effect from SNV alt allele perspective
dominant_isoform_pref_hap	Haplotype with higher dominant isoform usage
snv_astu_direction	+ if dominant isoform increased on SNV hap, − otherwise
snv_es_astu_signed	Signed ASTU effect from SNV alt allele perspective

event_snv.csv — Event-centric table

Column	Description
Sample, CellType	Sample and cell type identifiers
geneID, geneName, geneChr	Gene identifiers
gene_major_hap, es_ase, es_astu	Gene context (duplicated for self-containment)
ASE_call, ASTU_call	Final ASE / ASTU calls (3-category) for the gene, duplicated from gene_snv.csv — see that table for the rules.
dominant_isoform_overall, top_isoform_hap_major, top_isoform_hap_minor	Isoform context
eventID	Stable event key (event_type:start-end)
event_type	exon or junction
event_start, event_end	Event genomic coordinates
w_A_present, w_A_absent, w_B_present, w_B_absent	Weighted haplotype read counts
obs_hapA_include, obs_hapA_skip, obs_hapA_unobserved	Haplotype-A weighted read counts from raw BAM CIGAR observation: read alignment includes the event, splices over it (cassette skip / different junction), or fails to cover the event region (truncated).
obs_hapB_include, obs_hapB_skip, obs_hapB_unobserved	Same three categories for haplotype-B. The three columns sum to the per-read EM weight total in the joined pool (obs_* is per-read; existing w_* is isoform-multiplicity weighted, so the two are not equal when reads map to multiple isoforms).
obs_chi2, obs_p_value, obs_p_value_adj	Chi-square test on the 2×2 [[hapA_include, hapA_skip], [hapB_include, hapB_skip]] table — unobserved is dropped so truncated reads don't pollute the test. Runs whenever both row and column margins are non-zero (a single zero cell is kept — complete include/skip on one hap is the strongest allele-specific signal); only an all-zero row/column → insufficient_data. obs_p_value_adj = within-gene BH FDR across events where obs_test_type == 'chi2_hap_event'; NaN obs_p_value → NaN adj.
obs_test_type	chi2_hap_event (test ran), insufficient_data (2×2 sum < min_reads or a whole row/column margin is 0), or no_bam (per-gene read_blocks.pkl not present; other obs_* are None). Run --task step1_5 (per-sample SLURM array, one task per BAM) followed by --task step1_5_merge (single task) to populate the pkl cache.
event_inclusion_frac_A, event_inclusion_frac_B	Inclusion fraction per haplotype
event_pref_hap	Haplotype with higher event inclusion
event_pref_major_minor	major or minor relative to gene expression
event_chi2, event_p_value, event_p_value_adj	Haplotype-event association test (within-gene FDR), SCOTCH isoform-inferred membership. Compare against obs_* for sensitivity to read truncation.
has_linked_snv	Whether a nearby confident SNV is linked
linked_snv_count	Number of nearby SNVs linked to this event
is_nearest_snv_for_event	Whether this is the closest linked SNV
snvID, snv_pos, snv_ref, snv_alt	Linked SNV identity (NaN if none)
snv_hap, h_A, hat_Z_prob_revised	SNV phasing info (NaN if none)
exonic_distance, genomic_distance	Distance from SNV to event boundary
snv_expr_direction, snv_astu_direction	SNV regulatory interpretation
snv_event_direction	promotes_event or reduces_event
raw_validation_available	Whether raw read validation was performed
raw_ref_present, raw_ref_absent, raw_alt_present, raw_alt_absent	Raw BAM allele × event counts
raw_total_reads	Total raw reads in contingency table
raw_chi2, raw_p_value, raw_p_value_adj	Raw-read validation statistics. raw_chi2 is populated only for chi2_cross_event; for binomial_intra_event, raw_p_value is the binomial test result and raw_chi2 is None. raw_p_value_adj = within-gene BH FDR across all (event, SNV) raw tests; NaN raw_p_value → NaN adj.
raw_test_type	chi2_cross_event (default 2×2 chi-square) or binomial_intra_event (SNV inside the exon event — fallback binomial test on ref vs alt counts, p=0.5).

Canonical reference (with interpretation of the two event tests): docs/output_schema.md.

Configurable arguments

Required

Parameter	Description
--task	step5
--scotch_target	Path(s) to SCOTCH output directory
--bam_path	Aligned BAM file(s) (used for raw SNV–event chi-squared)
--output_folder	Output directory

Common optional

Parameter	Description	Default
--prefix	Output filename prefix	none
--cell_type_df_path	CSV with Cell / CellType columns	none
--gene_subset_path	Restrict to subset of genes	none
--n_workers	Parallel worker processes	1
--job_array_by_sample / --job_index	Sample-array execution	off / 0

Downstream knobs

Parameter	Description	Default
--event_min_reads	Minimum weighted read count per event test	10
--snv_event_distance	±bp exonic distance for SNV–event linking	50
--event_mode	all_events / switching_events / fdr_events	all_events
--fdr_events_value	FDR cutoff when event_mode=fdr_events	0.05
--astu_sig_only	Filter Task 4 to ASTU-significant genes (per cell type)	off
--astu_sig_from_bulk	Use Bulk ASTU significance for all cell types (overrides --astu_sig_only)	off
--astu_sig_threshold	p-value cutoff for ASTU gene filtering	0.05

Job completion

{output_folder}/job_markers/step5.done is written when step 5 finishes. To audit per-step missing genes across the whole pipeline, run --task check:

python src/longallele.py --task check \
    --scotch_target /path/to/scotch_output \
    --output_folder /path/to/results \
    --n_jobs 50

Citation

If you use LongAllele, please cite our preprint:

Xu Z, Wang K. LongAllele: a joint inference framework for allele-specific analysis on long-read bulk and single-cell RNA sequencing. bioRxiv 2026. https://doi.org/10.64898/2026.05.05.722992

@article{longallele2026,
  title   = {LongAllele: a joint inference framework for allele-specific
             analysis on long-read bulk and single-cell RNA sequencing},
  author  = {Xu, Zhuoran and Wang, Kai},
  journal = {bioRxiv},
  year    = {2026},
  doi     = {10.64898/2026.05.05.722992},
  url     = {https://www.biorxiv.org/content/10.64898/2026.05.05.722992}
}

Contributing and support

Bug reports, feature requests, and questions are welcome via GitHub Issues. Pull requests are also welcome — please open an issue first to discuss substantial changes.

License

LongAllele is released under the MIT License.