Data and scripts for longitudinal structural variant phylogenies in metastatic prostate cancer

This study evaluates whether structural variants (SVs) capture key evolutionary and resistance signals in mCRPC. We analyzed paired longitudinal tumor WGS (pre-BAT and on-BAT) with matched normals from COMBAT subjects treated with bipolar androgen therapy (BAT). SVs were called and genotyped, and SVCFit estimated structural-variant cellular fraction (SVCF), clustered SVs by prevalence, and inferred clonal phylogenies and clone proportions over time. Data show branched evolution and therapy-associated clonal shifts. In representative subjects, highly rearranged lineages contracted on BAT, while SV-defined clones expanded. Outputs include SV callsets, per-SV SVCF/CCF, cluster labels, trees, and figures for reuse.

This repository comprises the source data and computational workflows supporting "Longitudinal structural variant phylogenies define tumor evolution under therapeutic selection pressure in metastatic prostate cancer" by Liu et. al. (2026).

1. Clinical Cohort Data (COMBAT).
-- annotsv: Functional SV annotations generated via AnnotSV (Geoffroy et. al. 2018).
-- SVCFit_output: SV annotated with inferred SVCF, phasing, zygosity and overlapping CNVs (BED format).
-- cluster_result.csv: clusters of deduplicated SVs based on inferred SVCF.
-- sv2cluster.csv: clusters of all SVs based on inferred SVCF.
-- samp_pur.csv: Purity of all processed tumor samples estimated by FACETS (Shen et. al. 2016).
-- pair.bed: Matched pre- and on-BAT biopsy samples.
* The raw sequencing data and germline variant calls used in the current study are available from the corresponding author upon reasonable request. These are not publicly available due to informed consent restrictions.

2. Synthetic Benchmarking Data (VISOR_benchmark)
-- truth: Ground truth BED files defining spiked-in SVs: exact breakpoints , SV types (deletions, duplications, homozygosity).
-- depth: Test of the impact of variable breakpoint depth for SVCF estimation.
-- facet: CNV profiles generated via FACETS across five distinct experimental conditions to model heterogeneous tumor mixtures.
-- svclone_rdata: Serialized R objects containing subclonal deconvolution results (SVClone/ccube).
-- SV_vcfs: SV calls from Manta (Chen et. al. 2016).
-- SNP: SNP calls from GATK4 (McKenna et. al. 2010).

3. In Silico Validation (Prostate_mixture) Controlled admixtures of prostate cancer reads designed to evaluate sensitivity in polyclonal environments
-- prostate_vcfs: SV call sets for 3-, 4-, and 5-clone mixtures (3m, 4m, 5m) by Manta.
-- facet: CNV profiles generated via FACETS across all clone mixtures.
-- SNP: SNP calls from GATK4 across all clone mixtures.

4. Scripts
-- COMBAT: Clinical pipelines for alignment, consensus SV calling (integrating Manta, Delly, GRIDSS, and SURVIVOR), and functional annotation.
-- VISOR: Scripts for synthetic genome generation, read simulation, and benchmarking execution.
-- Prostate_mixture: Utilities for BAM mixing, interval splitting, and downstream processing.