SVhet is a pipeline for filtering heterozygous deletion calls in cohort-level VCFs using evidence from heterozygous sites in short-read sequencing data. It improves the reliability of heterozygous deletion calls by leveraging read-level and variant-level information across samples. SVhet does not filter other types of structural variants.

1. Features
2. Installation & Dependencies
3. Usage
4. Pipeline Overview
5. Output Format
6. Example Files & Run
7. Citation
8. Contact

• Filters heterozygous deletions based on genotype and quality metrics
• Per-sample read evidence extraction for wild-type (WT) and mutant (MUT) alleles
• Short variant calling on extracted read sets
• Heterozygosity evaluation within deletion regions and their flanking regions to flag unreliable calls
• Produces a single, annotated cohort-level VCF for downstream analysis

• bcftools
• bedtools
• pysam
• Python 3.6+
• numpy, tqdm

bash svhet.sh --ref <reference.fasta> \
              --sv-vcf <cohort.vcf.gz> \
              --outdir <output_dir> \
              --manifest <manifest.txt> \
              [--bed <regions.bed>] [--jobs <N>] [--keep-intermediate] [--min-dp <N>] [--high-het <N>]

Required Arguments

• --ref : Reference FASTA file
• --sv-vcf : Cohort-level SV VCF file (bgzipped and indexed)
• --outdir : Output directory
• --manifest : Tab-delimited file with sample ID (required), BAM path (required), and BAI path per line (optional)

Optional Arguments

• --bed : BED file of target regions
• --jobs : Number of parallel jobs (default: 1)
• --keep-intermediate : Keep intermediate files
• --min-dp : Minimum depth for reliable HETs (default: 5)
• --high-het : Minimum HET count to reject a DEL (default: 1)

The main entry point is svhet.sh, which orchestrates the following steps:

1. Generate SV Candidates (01_generate_candidates.sh)

   • Filters cohort VCF for deletion candidates with at least one heterozygous carrier.
   • Splits candidates by SV length (default: <1e6 and >1e6) and applies additional quality filters.
   • Optionally restricts to target regions using a BED file.

2. Extract Per-Sample Read Evidence (02_filter_by_samples.py)

   • For each sample and candidate, extracts WT and MUT supporting reads from the BAM file.
   • Writes these reads to temporary BAMs for downstream variant calling.
   • Handles both small and large SV candidates.

3. Short Variant Calling (03_call_variants.sh)

   • Calls short variants (SNPs/indels) on the WT and MUT BAMs using bcftools mpileup and bcftools call.
   • Filters for heterozygous sites.

4. Heterozygosity Evaluation (04_het_evaluator.py)

   • Compares the number of reliable heterozygous sites in WT and MUT callsets within each SV region.
   • Annotates the candidate VCF with the number of HETs and a filter status (PASS or HIGH_HET).
   • All sample-level VCFs are merged into a final, cohort-level annotated VCF.

After running the pipeline, a single bgzipped, annotated cohort-level VCF (VCFv4.2 format) is created (final-annotated.vcf.gz). This file contains SVhet-specific annotations for downstream filtering and interpretation.

• SVHET
  • PASS: Variant passes SVhet filtering
  • HIGH_HET: Variant flagged due to high heterozygosity in the region
• WT_HETS: Number of reliable heterozygous sites in the wild-type (WT) allele region (per sample)
• MUT_HETS: Number of reliable heterozygous sites in the mutant (MUT) allele region (per sample)

#CHROM	POS	ID	REF	ALT	QUAL	FILTER	INFO	FORMAT	SAMPLE1	SAMPLE2
chr1	123456	sv1	N	<DEL>	60	PASS	.	GT:WT_HETS:MUT_HETS:SVHET	0/1:3:0:PASS	0/0:0:0:PASS
chr2	234567	sv2	N	<DEL>	50	PASS	.	GT:WT_HETS:MUT_HETS:SVHET	0/1:5:2:HIGH_HET	0/1:4:0:PASS

In the example above, sv2 should be excluded from downstream analysis due to high heterozygosity detected from WT and MUT read evidences. For true heterozygous deletions, WT_HETS and MUT_HETS are typically 0 since only one haplotype exists in the deleted region. Currently, SVhet flags heterozygous deletions with >1 heterozygous site as HIGH_HET.

Example Manifest File

HG00096  /path/to/HG00096.bam /path/to/HG00096.bam.bai
HG00097  /path/to/HG00097.bam /path/to/HG00097.bam.bai

Notice there is an extra new line character in the end of file. The manifest file is tab-delimited.

Example Run

To run the test case, download the T2T reference from here. Decompress the reference gzip file and run SVhet as follows.

bash svhet.sh --ref chm13.v2.fasta \
             --sv-vcf test/chr1_127510500_128695280_HG00096.vcf.gz \
             --outdir test/results \
             --manifest test/manifest.txt \
             --jobs 4

Upon successful completion, the output file in test/results/final-annotated.vcf.gz should be the same as the one in test/output/. Use absolute paths if there is no output.

If you use SVhet in your research, please cite:

She, C.H., Chan, S.HS. & Yang, W. SVhet: towards accurate detection of germline heterozygous deletions using short reads. BMC Bioinformatics (2025). https://doi.org/10.1186/s12859-025-06342-7

For questions or issues, please contact Louis ([email protected]).