Choosing the right tool for protein structure prediction
3 Sep 2026
National Centre for Biomolecular Research
– Faculty of Science
Researchers affiliated with NCBR, Anna Hýsková, Eva Maršálková, and Petr Šimeček, have published a study comparing three widely used protein structure prediction tools: AlphaFold2, ESMFold, and OmegaFold. Using a benchmark dataset of 1 337 protein structures deposited between 2022 and 2024, they demonstrated that AlphaFold2 remains the most accurate method overall.
At the same time, the study shows that the much faster language model-based approaches, ESMFold and OmegaFold, often achieve comparable accuracy while requiring substantially fewer computational resources. The authors also identified specific cases in which alignment-free methods outperform AlphaFold2, particularly for de novo designed proteins. In addition, they developed machine learning models capable of predicting when the computational cost of AlphaFold2 is justified and when faster alternatives are sufficient.
Publication: Hýsková A., Maršálková E., Šimeček P. Balancing speed and precision in protein folding: a comparison of AlphaFold2, ESMFold, and OmegaFold. Frontiers in Genetics (2026). DOI: https://doi.org/10.3389/fgene.2025.1715037
