EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

MFMNet: Adaptive tri-layer graph pyramid with cross-scale attention router for synergistic drug prediction

Lei Wang, Liping Sun, Jingwen Pan, Guangming Shao, Guokun Shi, Guanyu Zhang, Yucheng Shi, Li Jin iD

PLOS One · 10.11.2026 · 21 (11) · e0362586

DOI: 10.1371/journal.pone.0362586 ↗

Kurz gefasst

Fachartikel in der Zeitschrift PLOS One, online erschienen am 10.11.2026. Verfasst von Lei Wang und 7 weiteren Autor:innen. Beteiligte Einrichtung: School of Medical Informatics Engineering, Anhui University of Chinese Medicine, Hefei, Anhui, China. Hauptthemen laut MEDLINE: Drug Synergism, Antineoplastic Agents, Computational Biology. Die Arbeit nennt 47 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Anhui Provincial Department of Education, Anhui Office of Philosophy and Social Science, Scientific Research Foundation of Education Department of Anhui Province of China.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background and objective Experimental screening of synergistic anticancer drug pairs is time-consuming, while effectively integrating complementary molecular graph representations with cell-line-specific information remains challenging for computational drug-synergy prediction.

Methods

We developed MFMNet with a Tri-Layer Graph Pyramid (TGP) comprising GCN, GIN, and GAT branches for extracting complementary molecular graph representations. A Cross-Scale Attention Router (CSAR), implemented as an adaptive feature-wise gating mechanism, integrates the graph representations before they are combined with drug embeddings and cell-line transcriptomic features for synergistic drug-pair prediction.

Results

In the initial five-fold cross-validation analysis, MFMNet achieved a mean five-fold AUROC of 0.9301 ± 0.0036. Under the stricter fold-matched evaluation protocol, the full MFMNet backbone achieved an AUROC of 0.9227 ± 0.0055 and a PR-AUC of 0.9127 ± 0.0084, with statistically significant improvements over XGBoost, DeepDDS-GCN, and DeepDDS-GAT after Holm correction.

Conclusions

MFMNet provides a computational framework for synergistic drug-pair prediction by integrating complementary molecular graph representations with drug and cell-line features. The results support its use for computational prioritization of candidate synergistic drug combinations. All reported performance estimates refer to record-level five-fold cross-validation within the O'Neil dataset; generalization to entirely unseen drugs and cell lines requires further study.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Drug SynergismAntineoplastic AgentsComputational BiologyHumansAlgorithmsGraph Neural Networks

Substanzen

Antineoplastic Agents

Autor:innen und Institutionen

  • School of Medical Informatics Engineering, Anhui University of Chinese Medicine, Hefei, Anhui, China.

ORCID-Profile: Li Jin

Förderung

  • Anhui Provincial Department of Education
  • Anhui Office of Philosophy and Social Science
  • Scientific Research Foundation of Education Department of Anhui Province of China

Zitieren

APA

Wang, L., Sun, L., Pan, J., Shao, G., Shi, G., Zhang, G., Shi, Y., & Jin, L. (2026). MFMNet: Adaptive tri-layer graph pyramid with cross-scale attention router for synergistic drug prediction. PLOS One, 21(11), e0362586. https://doi.org/10.1371/journal.pone.0362586

Vancouver

Wang L, Sun L, Pan J, Shao G, Shi G, Zhang G, et al. MFMNet: Adaptive tri-layer graph pyramid with cross-scale attention router for synergistic drug prediction. PLOS One. 2026;21(11):e0362586. doi:10.1371/journal.pone.0362586

BibTeX
@article{Wang2026,
  author = {Wang, Lei and Sun, Liping and Pan, Jingwen and Shao, Guangming and Shi, Guokun and Zhang, Guanyu and Shi, Yucheng and Jin, Li},
  title = {MFMNet: Adaptive tri-layer graph pyramid with cross-scale attention router for synergistic drug prediction},
  journal = {PLOS One},
  year = {2026},
  volume = {21},
  number = {11},
  pages = {e0362586},
  doi = {10.1371/journal.pone.0362586}
}
Kostenlos starten →