Researchers at the University of Colorado Anschutz (CU Anschutz) School of Medicine have completed one of the most exhaustive evaluations to date of software tools designed to analyze tandem repeats, a type of DNA variation that plays an important role in human disease and biological diversity.
Tandem repeats are stretches of DNA in which short sequences are repeated multiple times. While most are harmless, some expansions have been linked to neurological disorders, rare genetic diseases and other health conditions. Despite their importance, tandem repeats have historically been difficult to measure accurately.
Led by co-first authors Akshay Avvaru and Elbay Aliyev from the Dashnow Lab at the CU Anschutz Department of Biomedical Informatics, and senior author Harriet Dashnow, PhD, assistant professor of biomedical informatics, the study systematically evaluated leading computational tools used to identify and characterize tandem repeats from long-read DNA sequencing data.
"Short-read sequencing struggles with long or complex tandem repeats because the reads are simply too short to span them," said Dashnow. "With long-read sequencing now mainstream, we finally have the resolution to capture these regions, but there was a real bottleneck: researchers had a dozen new genotyping tools to choose from, yet no clear idea how they stacked up against each other in the real world. We needed to figure out which tools actually work before this data becomes standard in clinical and population studies."
Comparing the Tools Researchers Use
The team reviewed 25 bioinformatics methods and selected seven actively maintained tools for a large-scale benchmarking study.
Using publicly available Oxford Nanopore sequencing data from more than 100 individuals, the researchers evaluated performance across 43,009 tandem repeat locations throughout the human genome.
Rather than relying on a single measure of success, the team assessed each method using four complementary approaches:
- Concordance with high-quality reference genome assemblies
- Mendelian consistency within families
- Agreement across different tools
- Ability to detect disease-causing repeat expansions
This comprehensive approach allowed the researchers to examine not only whether a tool produced accurate repeat lengths, but also how it performed in situations relevant to research and clinical applications.
Accuracy Depends on the Challenge
The study found that most tools performed well when compared with reference genome assemblies, particularly when using newer Oxford Nanopore sequencing chemistry. However, performance was not consistent across all genomic scenarios.
Accuracy generally declined as repeat lengths increased. Many tools also struggled with:
- Very long repeat alleles
- Homopolymers, or repeats made of a single nucleotide
- Heterozygous sites where the two inherited copies differ
- Variants that differ substantially from the reference genome
One of the study's more surprising findings involved homopolymer stretches.
"We were really surprised by how much trouble tools had with homopolymer stretches," said Avvaru. "They accounted for almost half of all incorrect calls, even in the top-performing packages."
The findings highlight how complex tandem repeat analysis remains, even as sequencing technologies continue to improve.
Why Length Alone is Not Enough
One of the study’s most important findings was that commonly used performance measures did not fully predict a tool’s ability to identify known pathogenic repeat expansions.
Methods that performed well in assembly comparisons or Mendelian consistency analyses were not necessarily the most sensitive at detecting disease-associated expansions.
This suggests that measuring repeat length accurately is only part of the challenge.
"Measuring repeat length is only half the battle," said Aliyev. "Long-read sequencing lets us see the actual sequence inside the repeat, like subtle interruptions or specific disease-causing motifs. Catching these details can completely change how we predict disease risk, age of onset, and prognosis for patients, moving us past crude size estimates. However, not all tools detect these features with the same precision, and error rates and motif misclassifications can vary widely depending on the genotyper you choose."
According to the authors, sequence-level benchmarking provides a more complete picture of tool performance and may be essential when selecting methods for population-scale studies or future clinical applications.
A Roadmap for Future Development
Rather than identifying a single best-performing tool, the study found that different methods excelled in different areas.
“There's no silver bullet tool here,” said Aliyev. “If you want a solid, all-around performer for typical alleles and big expansions, LongTR is a safe bet. If your main goal is catching pathogenic expansions with high sensitivity, STRdust stands out.”
The results offer practical guidance for researchers selecting tandem repeat genotyping tools while also highlighting opportunities for future method development.
“Moving forward, we need to push sequence-level accuracy higher for really long alleles and clean up how tools model errors in homopolymers,” said Avvaru. "An important outcome of this study is that it has helped identify challenging regions and scenarios for ATaRVa, a genotyper developed by the lab of Divya Tej Sowpati at the Centre for Cellular and Molecular Biology (CCMB) in Hyderabad, India, and now being jointly developed in collaboration with our lab. We are currently working on optimizing our methods to improve genotyping accuracy and performance in these challenging regions."
As long-read sequencing becomes increasingly common in both research and clinical genomics, improving the accuracy and reliability of tandem repeat analysis will be critical for understanding genetic variation and identifying disease-causing mutations.