Abstract
High-throughput sequencing provides the means to determine the allelic decomposition for any gene of interest—the number of copies and the exact sequence content of each copy of a gene. Although many clinically and functionally important genes are highly polymorphic and have undergone structural alterations, no high-throughput sequencing data analysis tool has yet been designed to effectively solve the full allelic decomposition problem. Here we introduce a combinatorial optimization framework that successfully resolves this challenging problem, including for genes with structural alterations. We provide an associated computational tool Aldy that performs allelic decomposition of highly polymorphic, multi-copy genes through using whole or targeted genome sequencing data. For a large diverse sequencing data set, Aldy identifies multiple rare and novel alleles for several important pharmacogenes, significantly improving upon the accuracy and utility of current genotyping assays. As more data sets become available, we expect Aldy to become an essential component of genotyping toolkits.
Introduction: The Challenge of Complex Genotyping
While genotyping is a routine task for most of the genome, a critical subset of genes remains a significant computational challenge. Pharmacogenes, which determine individual responses to medications, are often both highly polymorphic and structurally variant. These genes frequently exist alongside nearly identical "pseudogenes," creating massive mapping ambiguity for short-read sequencing data.
Existing high-throughput sequencing (HTS) tools often fail to provide a complete "allelic decomposition"—the precise identification of the number of gene copies and their specific sequence content (star-alleles). This failure can lead to incorrect clinical interpretations and suboptimal personalized medicine.
The Aldy Framework
To address these challenges, the authors developed Aldy, a tool that frames genotyping as a combinatorial optimization problem. Aldy utilizes Integer Linear Programming (ILP) to find the most parsimonious explanation for sequencing data, effectively modeling gene fusions and copy number variations that other tools ignore.
The Aldy pipeline operates in four primary stages:
- Read Alignment & Mutation Detection: Standard alignment to identify potential variants.
- Structural Variation (SV) Detection: An ILP model estimates copy numbers by finding the simplest set of structures that explain the observed coverage.
- Major Star-Allele Identification: A second ILP model selects the most likely major alleles from a database that fit the detected structure.
- Genotype Refinement: Minor mutations are used to distinguish between sub-alleles, ensuring biological plausibility.
Results and Validation
Aldy was evaluated using diverse datasets, including PGRNseq targeted capture, Illumina Whole-Genome Sequencing (WGS), and GeT-RM reference materials.
Superior Accuracy in Complex Regions
On CYP2D6, one of the most difficult pharmacogenes to genotype due to its high structural variability, Aldy significantly outperformed existing tools like Cypiripi and Astrolabe, which failed in 40–50% of cases where Aldy succeeded.
Discovery of Novel Alleles
Beyond matching known genotypes, Aldy identified several previously uncharacterized major star-alleles in the 1000 Genomes and GeT-RM datasets. These included novel variants of CYP2D6, DPYD, and CYP2C19, demonstrating the tool's utility in expanding our understanding of genetic diversity.
Mendelian Consistency and Efficiency
In family trio studies, Aldy’s results were perfectly consistent with Mendelian inheritance laws, a feat often missed by SNP-based methods. Furthermore, the tool is highly efficient, typically processing a sample in under 10 seconds with minimal memory overhead, making it suitable for high-throughput clinical pipelines.
Discussion and Future Directions
While Aldy represents a major leap forward, it is currently constrained by the limitations of short-read sequencing, particularly in "phasing" mutations that are separated by distances greater than a single read length. Additionally, its structural detection relies on a database of known fusion types.
Future iterations of the tool may incorporate long-read sequencing (e.g., PacBio or Oxford Nanopore) to resolve these phasing ambiguities and identify de novo structural breakpoints without database reliance. As genomic data continues to scale, Aldy provides a robust foundation for the precise genotyping required for truly personalized medicine.
Summary updated