[Paper Review] grenedalf: population genetic statistics for the next generation of pool sequencing

Lucas Czech, Jeffrey P. Spence+1 more|arXiv (Cornell University)|Jun 20, 2023

Genomics and Phylogenetic Studies4 citations

TL;DR

This paper introduces grenedalf, a high-performance C++ command-line tool that computes key population genetic statistics—such as nucleotide diversity (θ), Tajima’s D, and F ST—for pool-sequencing data. It addresses limitations in existing tools by offering orders-of-magnitude faster processing, support for diverse input formats, and unbiased estimators that correct for sampling noise in low-coverage and small-pool sequencing experiments.

ABSTRACT

Pool sequencing is an efficient method for capturing genome-wide allele frequencies from multiple individuals, with broad applications such as studying adaptation in Evolve-and-Resequence experiments, monitoring of genetic diversity in wild populations, and genotype-to-phenotype mapping. Here, we present grenedalf, a command line tool written in C++ that implements common population genetic statistics such as $θ$, Tajima's D, and FST for Pool sequencing. It is orders of magnitude faster than current tools, and is focused on providing usability and scalability, while also offering a plethora of input file formats and convenience options.

Motivation & Objective

• To address the lack of scalable, user-friendly, and statistically robust tools for analyzing pool-sequencing data in population genetics.
• To correct for biases in allele frequency estimation arising from finite sampling of individuals and reads in pool-seq experiments.
• To enable efficient computation of core statistics like θ, Tajima’s D, and F ST across large datasets with thousands of samples.
• To support modern pool-seq pipelines by integrating with frequency-based inputs, including those from HAF-pipe and other HARP-based methods.
• To improve usability through modular design, multi-format input support, and convenience tools for data filtering and merging.

Proposed method

• Re-derives and implements unbiased estimators for nucleotide diversity (π), Watterson’s θ, Tajima’s D, and Nei’s and Hudson’s F ST, accounting for pool-sequencing noise.
• Uses a unified Variant data type to represent genomic positions with reference/alternative alleles and sample-specific base counts, abstracting over input file formats.
• Employs a Variant Iterator that streams and synchronizes multiple input files (e.g., SAM, mpileup, VCF, tabular) in parallel, enabling multi-threaded processing.
• Supports window-based analysis and direct computation of statistics on streamed genomic positions, with optional buffering for performance.
• Builds on the high-performance genesis C++ library, ensuring optimal speed and modularity for future extension with new file formats or statistics.
• Provides a Python binding roadmap to enhance accessibility for bioinformatics developers and integrate with existing analysis workflows.

Experimental results

Research questions

• RQ1 How can population genetic statistics be accurately estimated in pool-sequencing data, where sampling noise from finite pools and low-coverage sequencing introduces bias?
• RQ2 What are the statistical shortcomings of existing tools like PoPoolation and poolfstat in estimating Tajima’s D and F ST, particularly for small sample sizes and low coverage?
• RQ3 Can a modular, high-performance C++ tool be designed to support diverse input formats while maintaining accuracy and usability for large-scale pool-seq studies?
• RQ4 How can allele frequency data from HAF-pipe and other HARP-based pipelines be directly used for population genetic analysis without requiring raw read counts?
• RQ5 What performance gains and scalability improvements can be achieved through a modern, optimized C++ implementation compared to legacy tools written in Perl, R, or C?

Key findings

• The authors identify and correct long-standing biases in the Tajima’s D estimator previously implemented in PoPoolation, which were not resolved by subsequent bug fixes.
• The F ST estimators in PoPoolation2 (Kofler and Karlsson) are shown to be systematically biased upward under low-coverage and small-pool conditions.
• The new unbiased estimators for π_within, π_between, and π_total lead to more accurate F ST estimates, especially in low-coverage and small-sample pool-sequencing experiments.
• grenedalf achieves orders-of-magnitude speedup over existing tools, enabling efficient analysis of datasets with thousands of pool-sequencing samples.
• The software supports a wide range of input formats—including SAM, mpileup, VCF, and tabular frequency files—through a unified, extensible Variant abstraction.
• The tool can directly process pre-computed allele frequencies, such as those from HAF-pipe, enabling seamless integration with modern pool-seq pipelines and reducing computational overhead.