Repository for "Evaluating local assembly strategies for long-read structural variant valling"

Scripts, code and Snakemake workflow for simulation of SVs on T2T-CHM13 and ONT long reads for evaluation of different ways to generate properly phased local consensus sequences at SV loci.

Project structure

Scripts for data generation (python) and evaluation (R) are located in scripts/.

C++ source code for extraction of region sequences* from alignment data, creation of consensus sequences with different approaches and alignment of consensus sequences against true allele sequences for all variants is located in src/. Code can be compiled with make all. Dependencies are SPOA and seqan2.

A conda environment for compilation can be installed with

conda create -n lr_benchmark_env -c conda-forge -c bioconda htslib=1.21 seqan=2.4 spoa=4.1.5 cxx-compiler libdeflate=1.22 zlib rapidfuzz-cpp abpoa=1.5.6

Results

Plots underlying the Figures in the paper were generated with scripts/create_paper_plots.R and can be found in results/plots. Supplementary Plots were generated with scripts/create_supplementary_figures.R and can be found in results/supplementary_plots. Data shown in the plots or mentioned in the paper can be found as tables in results/numbers.

Relevant data for the plot generation can be found in:

• results/ont and results/pacbio, divided in subfolders containig output of ./evaluate_consensus sequences: tables with results of consensus sequence alignments to true haplotype sequences
• data/time_data, containing tables with statistics about consensus sequence generation methods, including runtime for each locus

Simulated data and generated consensus sequences are not contained in this repository.

Workflow

1. Insertion of 16000 SVs on chromosome 1 of T2T-CHM13.
   • Deletions
   • Inversions
   • Tandem Duplications
   • Compound Heterozygous Deletions
   • Compound Heterozygous Inv + Del (no overlap) Simulated sizes: 50, 200, 500, 1000
2. Simulation of long read data using PBSIM
   • ONT, PacBio HiFi
   • 20x, 30x, 60x coverage
   • Accuracy 99%
3. Alignment with Minimap2
4. Extraction of sequences spanning the variant regions from long read alignments
5. Creation of consensus sequence(s) using 6 different methods
   • Heaviest Bundle Algorithm (abPOA)
   • K-Mer based clustering + POA (SPOA)
   • Iterative generation of multiple POA graphs (SPOA)
   • A priori phasing
   • RapidFuzz
   • Based on haplotype of origin (known through simulation)
6. Alignment of consensus sequences against true allele sequences for each variant