Fine-tuning a sentence transformer for DNA
The findings indicated that the refined proposed model generated DNA embeddings that exceeded DNABERT in multiple tasks. However, the proposed model was not superior to the nucleotide transformer in terms of raw classification accuracy. The nucleotide transformer excelled in most tasks; but, this superiority incurred significant computing expenses, rendering it impractical for resource-constrained environments such as low- and middle-income countries (LMICs). The nucleotide transformer also performed worse on retrieval tasks and embedding extraction time. Consequently, the proposed model presents a viable option that balances performance and accuracy.
Authors
- Riana Bornman (ORCID: https://orcid.org/0000-0003-3975-2333)
- Vanessa M. Hayes (ORCID: https://orcid.org/0000-0002-4524-7280)
- Mpho Mokoatle (ORCID: https://orcid.org/0000-0001-7689-2840)
- Vukosi N. Marivate (ORCID: https://orcid.org/0000-0002-6731-6267)
- Darlington Mapiye (ORCID: https://orcid.org/0000-0001-8403-8783)
Institutions
- AstraZeneca (United Kingdom) (GB)
- The University of Sydney (AU)
- University of Manchester (GB)
- Cancer Research UK Manchester Institute (GB)
- University of Pretoria (ZA)
Publication Details
- Journal
- BMC Bioinformatics
- Published
- 2025-10-29
- DOI
- https://doi.org/10.1186/s12859-025-06291-1
- Primary Topic
- Machine Learning in Bioinformatics
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- University of New South Wales
- Center for High Performance Computing