Caveat emptor: predicting and modeling protein–DNA recognition and binding via machine-learning computational approaches
Nucleic Acids Research · 2026
DOI: 10.1093/nar/gkag608 ↗Kurz gefasst
Übersichtsarbeit in der Zeitschrift Nucleic Acids Research, 2026. Verfasst von Morgan A Esler und 9 weiteren Autor:innen. Beteiligte Einrichtung: Division of Basic Sciences, Fred Hutchinson Cancer Center, 1100 Fairview Ave. N., Seattle, WA 98109, United States. Hauptthemen laut MEDLINE: Computational Biology, DNA, DNA-Binding Proteins, Machine Learning. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 58 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: NIH HHS, NIGMS NIH HHS, Fred Hutchinson Cancer Center.
Abstract (Originalsprache)
The recent development of AI-based predictive tools, such as AlphaFold3, for the prediction of the structures of biological molecules and their complexes has transformed modern molecular and cellular biology. While it displays exceptional accuracy in the modeling of folded protein domains and subunits, as well as larger protein–protein complexes and assemblages, AlphaFold3’s performance in predicting the details of protein–DNA (or more broadly, protein–nucleic acid) contacts and complexes is less well established. Here we summarize the recent development and performance of tools intended to predict, model, and/or design protein:DNA recognition and contacts, and then demonstrate (using a well-defined system that offers a minimal “degree of difficulty”) the issues that often surround the use of a resource such as AlphaFold3 for predicting protein:DNA interactions. Beyond providing a cautionary tale for casual users, we note that the incorporation of hybrid models of protein–DNA complexes (in which computationally predicted models are docked into low-resolution CryoEM density maps with little further refinement or quality control) into future training sets may lead to an ongoing and inappropriate learning cycle that further encourages such tools to generate new, equally inaccurate models of protein–DNA complexes.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Substanzen
Autor:innen und Institutionen
- Division of Basic Sciences, Fred Hutchinson Cancer Center, 1100 Fairview Ave. N., Seattle, WA 98109, United States.
ORCID-Profile: Rachel Werther, Lindsey A Doyle, Barry L Stoddard
Förderung
- NIH HHS
- NIGMS NIH HHS
- Fred Hutchinson Cancer Center
- Advanced Light Source
- Department of Energy Office of Science User Facility
- National Institutes of Health
- National Institute of General Medical Sciences
Zitieren
APA
Esler, M. A., Werther, R., Doyle, L. A., Ubilla-Rodriguez, N. C., Schwensen, J. S., Hallinan, J. P., Lambert, A. R., Young, J. C., Silverstein, M., & Stoddard, B. L. (2026). Caveat emptor: predicting and modeling protein–DNA recognition and binding via machine-learning computational approaches. Nucleic Acids Research. https://doi.org/10.1093/nar/gkag608
Vancouver
Esler MA, Werther R, Doyle LA, Ubilla-Rodriguez NC, Schwensen JS, Hallinan JP, et al. Caveat emptor: predicting and modeling protein–DNA recognition and binding via machine-learning computational approaches. Nucleic Acids Research. 2026. doi:10.1093/nar/gkag608
BibTeX
@article{Esler2026,
author = {Esler, Morgan A and Werther, Rachel and Doyle, Lindsey A and Ubilla-Rodriguez, Natalia C and Schwensen, Jeanette S and Hallinan, Jazmine P and Lambert, Abigail R and Young, Juliana C and Silverstein, Miriam and Stoddard, Barry L},
title = {Caveat emptor: predicting and modeling protein–DNA recognition and binding via machine-learning computational approaches},
journal = {Nucleic Acids Research},
year = {2026},
doi = {10.1093/nar/gkag608}
}
EUVIMED