Impute-First Alignment framework

Introduction

The Impute-First alignment framework aims to minimize reference bias by creating personalized references. This method enhances variant-calling accuracy, precision, and efficiency in whole-genome DNA sequencing analysis.

Workflow Overview

Figure 1: The Impute-first alignment workflow

Repository Structure

• imputefirst_workflows/upstream_personalization/: Snakemake workflow to generate personalized diploid references from downsampled reads by performing genotyping and imputation using a reference panel.

• preprocessing_upstream_pipelines/: Workflow assessing different genotyping and imputation method combinations for generating personalized references.

• imputefirst_workflows/downstream_analysis_with_giraffe/: Downstream analysis using Impute-First personalized diploid references of a given sample, built into variation graphs, enabling alignment and lift-over with VG Giraffe (Figure 1, Section B.1) and to perform variant calling.

• imputefirst_workflows/downstream_analysis_with_leviosam2/: Downstream analysis using Impute-First personalized diploid FASTA references of a given sample, aligned with BWA-MEM and lifted with LevioSAM2 to perform variant calling (Figure 1, Section B.2).

• benchmarking_downstream_workflows/: Scripts to generate GRCh38-aligned BAMs for benchmarking Impute-First personalized diploid references against linear, pangenome, personalized-pangenome, and ground-truth reference combinations for a given sample.

• evaluation/: Scripts for computing evaluation metrics:

  • Genotyping and Imputation Calls evaluation: Call Accuracy, Window Accuracy.
  • Downstream analysis evaluation: Allelic Balance at HETs, Variant Call Accuracy.

Running the Upstream Personalization Workflow

The upstream personalization workflow is implemented using Snakemake and defined in upstream_personalization/. It performs read downsampling, genotyping, and imputation to generate diploid personalized reference for a given sample.

All required software dependencies are listed in the Conda environment file env.yml. A compatible environment can be created with:

conda env create -f imputefirst_workflows/upstream_personalization/env.yml
conda activate genotyping_imputation

To test the workflow, a chromosome 21–based demonstration is provided in the upstream_personalization/ directory. It includes the following files:

• download_exampleData.sh: Downloads a small test dataset (HG002 chr21 reads and reference files).
• download_linkage_maps.sh: Downloads linkage maps for use with Beagle and GLIMPSE.
• configs/exampleData_config.yaml: Configuration file for running the workflow on the example dataset.

Run the workflow with:

snakemake -j <threads> --configfile configs/exampleData_config.yaml

This config file parameters can be modified to enable using different input files.

Dependencies and setup instructions for downstream and benchmarking modules are described in their respective subdirectories.

Running the Downstream Analysis Workflow

LevioSAM2 based workflow

VG giraffe based workflow

Getting Started

1. Clone the repository to your local machine.
2. Navigate to each module's directory and follow the instructions in the README.md files to execute the scripts.

Contact

For inquiries or support related to the Impute-First alignment framework, please reach out to us.

• Naga Sai Kavya Vaddadi - kvaddad1@jhu.edu
• Ben Langmead - langmea@cs.jhu.edu

Johns Hopkins University, Department of Computer Science

Acknowledgments

This work was generously supported by NIH Grant R01HG011392. Special thanks to Langmead lab colleagues.

Available Resources

preprint: Vaddadi, Naga Sai Kavya, Taher Mun, and Ben Langmead. "Minimizing Reference Bias with an Impute-First Approach." bioRxiv (2023): 2023-11.

abstract: Vaddadi, Naga Sai Kavya, Taher Mun, and Benjamin Langmead. "Minimizing Reference Bias: The Impute-First Approach for Personalized Genome Analysis." Proceedings of the 14th ACM International Conference on Bioinformatics, Computational Biology, and Health Informatics. 2023.

CSHL Genome Informatics Talk 2023, Slides: Minimizing reference bias with an impute-first approach.