EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Multi-Property De Novo Drug Design Using Deep Learning-Based Knowledge Distillation and Reinforcement Learning

Liuying Wang, Zhao Lu, Lijuan Cui, Chang Liu, Yuting Qin, Shundan Feng, Dongxue Wang, Weixue Yin, Zheng Kang, Lei Cao

International Journal of Molecular Sciences · 2026

DOI: 10.3390/ijms27146125 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Molecular Sciences, 2026. Verfasst von Liuying Wang und 9 weiteren Autor:innen. Beteiligte Einrichtung: School of Health Management, Harbin Medical University, Harbin 150086, China. Hauptthemen laut MEDLINE: Deep Learning, Drug Design. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 34 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Education Department of Heilongjiang Province, Heilongjiang Provincial Metrology and Testing Institute.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Computer-aided de novo drug design has been widely explored for early-stage drug discovery, yet the multi-property optimization of novel molecules remains challenging. We aimed to develop a de novo drug design model to efficiently optimize multiple properties simultaneously. We developed a teacher–student-interaction deep learning model fine-tuned by reinforcement learning (TSItransRL) using bioactivity datasets (DRD2 and JNK3/GSK3β targets). A conditional transformer was pretrained as the teacher model to incorporate multi-property information. A vanilla transformer served as the student model and was subsequently optimized through interactive knowledge distillation and reinforcement learning. An evaluation was conducted using MOSES and conditional metrics on two tasks, specifically generating molecules with DRD2-targeting activity and generating molecules with dual JNK3/GSK3β-targeting activity, with the analyses including docking, the similarity ensemble approach (SEA), and scaffold novelty. TSItransRL achieved success rates of 98.36% and 98.90% for the DRD2 and JNK3/GSK3β tasks, respectively, with an internal diversity of 0.795, outperforming most baselines. The docking, SEA, scaffold, and ADMET analyses were used as exploratory in silico assessments to support the preliminary prioritization of selected generated molecules. TSItransRL provides an in silico framework for benchmark-level multi-property molecular generation and prioritization, combining interactive knowledge distillation with reinforcement learning to explore molecules that satisfy predefined predicted-activity, drug-likeness, and synthetic-accessibility criteria. The generated molecules should be regarded as computational candidates for a further medicinal-chemistry assessment, independent validation, and experimental testing rather than experimentally validated leads.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Deep LearningDrug DesignReinforcement Machine LearningGlycogen Synthase Kinase 3 betaMolecular Docking SimulationHumansMitogen-Activated Protein Kinase 10Receptors, Dopamine D2

Stichwörter der Autor:innen

computer-aided drug designde novo drug designdeep learningmulti-property optimizationreinforcement learning

Substanzen

Glycogen Synthase Kinase 3 betaMitogen-Activated Protein Kinase 10Receptors, Dopamine D2

Autor:innen und Institutionen

  • School of Health Management, Harbin Medical University, Harbin 150086, China.
  • Department of Biostatistics, School of Public Health, Harbin Medical University, Harbin 150086, China.

Förderung

  • National Natural Science Foundation of China
  • Education Department of Heilongjiang Province
  • Heilongjiang Provincial Metrology and Testing Institute

Zitieren

APA

Wang, L., Lu, Z., Cui, L., Liu, C., Qin, Y., Feng, S., Wang, D., Yin, W., Kang, Z., & Cao, L. (2026). Multi-Property De Novo Drug Design Using Deep Learning-Based Knowledge Distillation and Reinforcement Learning. International Journal of Molecular Sciences. https://doi.org/10.3390/ijms27146125

Vancouver

Wang L, Lu Z, Cui L, Liu C, Qin Y, Feng S, et al. Multi-Property De Novo Drug Design Using Deep Learning-Based Knowledge Distillation and Reinforcement Learning. International Journal of Molecular Sciences. 2026. doi:10.3390/ijms27146125

BibTeX
@article{Wang2026,
  author = {Wang, Liuying and Lu, Zhao and Cui, Lijuan and Liu, Chang and Qin, Yuting and Feng, Shundan and Wang, Dongxue and Yin, Weixue and Kang, Zheng and Cao, Lei},
  title = {Multi-Property De Novo Drug Design Using Deep Learning-Based Knowledge Distillation and Reinforcement Learning},
  journal = {International Journal of Molecular Sciences},
  year = {2026},
  doi = {10.3390/ijms27146125}
}
Kostenlos starten →