Learning the assay, not the hazard: how computational toxicity models inherit the artifacts of their high-throughput training data
Archives of Toxicology · 2026-12 · 100 (12) · 5703-5706
DOI: 10.1007/s00204-026-04512-x ↗Kurz gefasst
Leserbrief in der Zeitschrift Archives of Toxicology, online erschienen am 22.07.2026. Verfasst von Muhammad Javid Iqbal und 2 weiteren Autor:innen. Beteiligte Einrichtung: Doctoral Program in Sciences, Specialization in Applied Cellular and Molecular Biology, Universidad de La Frontera, 4811230, Temuco, Chile. m.iqbal01…. Hauptthemen laut MEDLINE: High-Throughput Screening Assays, Computer Simulation, Toxicity Tests, Toxicology. Die Arbeit nennt 9 Literaturangaben.
Abstract (Originalsprache)
High-throughput screening data have become the de facto ground truth for in silico toxicology. But a high-throughput “active” can reflect target engagement, non-specific cytotoxicity, or assay interference and the programmes that generate these data flag the latter two with dedicated counter-screens. When models are trained on the hit-calls without those flags, they can learn how the assay behaved rather than how the chemical harms. Because the confounders are structurally systematic, they are exactly the kind of signal a structure-based model will capture. This Commentary sets out the problem and four low-cost controls.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Doctoral Program in Sciences, Specialization in Applied Cellular and Molecular Biology, Universidad de La Frontera, 4811230, Temuco, Chile. m.iqbal01@ufromail.cl.
- Laboratory of Natural Products & Drug Discovery, Department of Basic Sciences, Faculty of Medicine, Universidad de La Frontera, 4780000, Temuco, Chile. m.iqbal01@ufromail.cl.
- Faculty of Veterinary Sciences, University of Veterinary and Animal Sciences, Lahore, 54000, Pakistan.
- Laboratory of Natural Products & Drug Discovery, Department of Basic Sciences, Faculty of Medicine, Universidad de La Frontera, 4780000, Temuco, Chile. cristian.paz@ufrontera.cl.
ORCID-Profile: Muhammad Javid Iqbal
Zitieren
APA
Iqbal, M. J., Amjad, T., & Paz, C. (2026). Learning the assay, not the hazard: how computational toxicity models inherit the artifacts of their high-throughput training data. Archives of Toxicology, 100(12), 5703-5706. https://doi.org/10.1007/s00204-026-04512-x
Vancouver
Iqbal MJ, Amjad T, Paz C. Learning the assay, not the hazard: how computational toxicity models inherit the artifacts of their high-throughput training data. Archives of Toxicology. 2026;100(12):5703-5706. doi:10.1007/s00204-026-04512-x
BibTeX
@article{Iqbal2026,
author = {Iqbal, Muhammad Javid and Amjad, Tooba and Paz, Cristian},
title = {Learning the assay, not the hazard: how computational toxicity models inherit the artifacts of their high-throughput training data},
journal = {Archives of Toxicology},
year = {2026},
volume = {100},
number = {12},
pages = {5703--5706},
doi = {10.1007/s00204-026-04512-x}
}
EUVIMED