Back to Home

Study finds duplicate genomes with taxonomic conflicts in a widely used genomic database

Bharti Dharapuram
30 Oct 2026
research-bytes

A recent study identified duplicate genome sequences across multiple records in the NCBI database, many of which carried conflicting taxonomic labels. The image above shows the distribution of label inconsistencies across different taxonomic levels; ‘No difference’ refers to duplicate genomes with matching taxonomic labels. The size of the bubble and the number within refer to the number of sets of duplicate genomes at a given taxonomic level. The number in parentheses gives the total number of genomes across these sets. Image: [Saraf & Sharma (2026)](https://onlinelibrary.wiley.com/doi/10.1002/imo2.70140), Wikimedia Commons (CC BY-SA 4.0)

Millions of genome sequences available in publicly accessible databases are widely used in biological research. As sequencing technology becomes cheaper and faster, these databases continue to expand rapidly. However, a recent study has uncovered errors in these databases, with thousands of instances where identical genome sequences appear across multiple records, often with conflicting taxonomic labels. Left unresolved, these errors could creep into downstream research, ranging from pathogen surveillance and antimicrobial resistance to species diversity and evolutionary relationships.


The study, carried out by Niharika Saraf and Dr Gaurav Sharma from the Department of Biotechnology, IIT Hyderabad, analysed 729,559 prokaryote genomes in the National Center for Biotechnology Information (NCBI) genome database. The database hosts genome sequences submitted by researchers from around the world, which are freely available to anyone interested in using them. The researchers identified 3334 sets of duplicate genomes, many of them with inconsistent taxonomic labels ranging from the superkingdom level to strains within a species. The study highlights the limitations of quality checks in the existing genome submission process and recommends screening procedures to reduce errors.

The idea for the study came about when the authors noticed two identical genomes assigned to different genera in a dataset they were analysing.

The idea for the study came about when the authors were working on a project comparing the genomes of ‘fruiting gliding’ bacteria from the phylum Myxococcota. To their surprise, they noticed two genomes with identical sequences assigned to different genera in their dataset. “This observation prompted us to extend the investigation to all prokaryotic genomes available at NCBI to systematically assess the prevalence of sequence-identical genomes,” says Niharika, an IITH doctoral student and co-corresponding author of the study. “It reminded us of the importance of keeping an eye on our datasets, as they might reveal something important.”


A closer look at large genome datasets


To dig deeper into how pervasive these errors are, the researchers used a bioinformatics pipeline to screen all prokaryotic genome records with identical size, base composition and the number of contigs (contiguous stretches of sequence in the genome assembly). From these, they zeroed in on records with 100% sequence identity, where the same genome sequences occurred under multiple records. This indicates potential error, as it is unlikely for even individuals of the same species to have the exact same genome sequence.


The researchers identified 3334 sets of duplicated genomes, most comprising two different records with identical genome sequences. Most duplicates showed discordant taxonomy at the strain level (2250 sets, 4571 records), followed by a significant number lacking complete taxonomic information (396 sets, 793 records). Looking more closely, the researchers found instances where duplicated genomes shared some details related to submitting institution, sampling information and sequence technology, but differed in other metadata, indicating potential errors during submission. Other cases where identical sequences were assigned to related species suggest inaccurate taxonomic identification carried out using unsuitable or incomplete reference databases.

The researchers identified 3334 sets of duplicated prokaryote genomes in the NCBI database, many comprising two different records with identical sequences assigned to different strains.

While NCBI does have checks in place, its use of a metric of genomic similarity called average nucleotide identity to screen for duplication may not capture sequence identity across the entire genome. Additionally, mismatches in classification systems between records, and poor synchronization/updation of taxonomic information also lead to identification errors. To address these challenges, the researchers propose various measures to address duplicate genome sequences, including using better measures of sequence identity, linking records to taxonomic classification systems, and cross-referencing them with more well-curated taxonomic databases.


Data reliability in the age of AI


Public genomic databases offer a valuable shared resource, but may carry hidden errors, which affect the reliability of the data and any downstream analysis. “Our experience highlights the value of systematically examining large public datasets, as inconsistencies that may go unnoticed in individual studies can become apparent when genomes are compared at scale”, says Dr Gaurav. “Our findings have made us more cautious about treating publicly available genome records as independent biological entities based solely on their taxonomic annotations,” adds Niharika. They recommend a bit-by-bit comparison of files alongside metadata and taxonomy checks to improve the reliability and reproducibility of large-scale comparative genomics analysis.

“We are entering an era where generating data is becoming easier than validating it. The real challenge is in ensuring that the insights generated by AI are grounded in reliable data.”

“We are entering an era where generating data is becoming easier than validating it. AI will undoubtedly accelerate how we analyse, interpret and extract knowledge from increasingly massive biological datasets,” Dr Gaurav says, reflecting on data-driven biology in the age of AI. “However, the real challenge will be ensuring that the insights generated by AI are grounded in reliable data. The future of scientific research will depend not just on how intelligently we analyse data, but on how carefully we validate it in the first place.”


Reference: Saraf, N., & Sharma, G. (2026). Sequence-identical microbial genomes with disparate taxonomy: A large-scale redundancy analysis in NCBI genome database. iMetaOmics, e70140. https://doi.org/10.1002/imo2.70140

Biotechnology
#genomics #genomic databases #data quality #data-driven biology #open science