AI-guided De Novo design of DHPS inhibitors.
- 1Department of Biotechnology, College of Life Sciences and Biotechnology, Korea University, Seoul, 02841, Republic of Korea.
- 2Department of Biomedical Sciences, College of Medicine, Korea University, Seoul, 02841, Republic of Korea. [email protected].
- 3Department of Biotechnology, College of Life Sciences and Biotechnology, Korea University, Seoul, 02841, Republic of Korea. [email protected].
Abstract
Hypusine modification of eIF5A by Deoxyhypusine Synthase (DHPS) is essential for the replication of a broad range of viruses, making DHPS an attractive host-directed therapy (HDT) target for antiviral drug discovery. AI-based drug design methods spanning binding affinity prediction and de novo molecular generation are rapidly proliferating, yet systematic benchmarking of diverse algorithmic paradigms under identical conditions on a shared target, with experimental validation, remains scarce. Here, we present an integrated AI benchmarking pipeline targeting DHPS and evaluate the full process from computational prediction to experimental validation. We compared 13 MM-GBSA score estimation models and 10 de novo molecular generation methods under 5-seed repeated experiments. For the prediction task, we propose an Out-of-Fold stacking ensemble integrating gradient boosting, graph neural networks, and chemical language models, achieving the highest Spearman rank correlation of 0.861 among all compared models. For the generation task, we propose RL-Design, a reinforcement learning-based framework using the ensemble predictor as a scoring oracle. RL-Design achieved the highest hit rate of 92.02% and maintained drug-like physicochemical properties across generated molecules, outperforming all nine competing methods. Moreover, experimental dose-response assays confirmed inhibitory activity in all three predicted-active compounds, with two highly active compounds showing IC₅₀ values of 31.2 and 49.1 µM. Predicted and experimental activity showed complete concordance across all six tested compounds, providing oracle-independent validation of the pipeline. This pipeline provides a reproducible framework for AI-guided, target-specific inhibitor design and demonstrates the feasibility of HDT-focused antiviral drug discovery through systematic AI benchmarking.Scientific contributionThis work proposes two methods for inhibitor discovery against DHPS, a host-directed antiviral target. A predictive model for MM-GBSA binding-score estimation and a generative model that designs candidate hit compounds both ranked first among previous prediction and generation models under identical benchmarking. The predicted activities were experimentally confirmed by dose-response enzymatic assays.
Related
- DEG2MOL: Conditional Latent Flow Matching for Transcriptome-Guided de novo Drug Design
- CADD-guided design of novel triclosan–hydrazone hybrids with antibacterial and antimycobacterial activity as InhA inhibitors
- De novo Design of a Vanadium Haloperoxidase
- Machine-learning-guided discovery of selective CK2 inhibitors with reduced CDK off-target activity
- Pharmacophore-Conditioned Diffusion Model for Ligand-Based De Novo Drug Design
BibTeX
@article{shin2026aiguided,
title = {{AI-guided De Novo design of DHPS inhibitors.}},
author = {Hongchul Shin and Kyuhyeon Bang and Ingyo Park and Hong-Rae Kim and Kwang Yeon Hwang},
journal = {Journal of cheminformatics},
year = {2026},
month = aug,
doi = {10.1186/s13321-026-01277-4},
url = {https://europepmc.org/article/MED/42968194},
note = {CC BY-NC-ND},
}
RIS
TY - JOUR
TI - AI-guided De Novo design of DHPS inhibitors.
AU - Hongchul Shin
AU - Kyuhyeon Bang
AU - Ingyo Park
AU - Hong-Rae Kim
AU - Kwang Yeon Hwang
PY - 2026
DA - 2026/08/05
JO - Journal of cheminformatics
DO - 10.1186/s13321-026-01277-4
UR - https://europepmc.org/article/MED/42968194
AB - Hypusine modification of eIF5A by Deoxyhypusine Synthase (DHPS) is essential for the replication of a broad range of viruses, making DHPS an attractive host-directed therapy (HDT) target for antiviral drug discovery. AI-based drug design methods spanning binding affinity prediction and de novo molecular generation are rapidly proliferating, yet systematic benchmarking of diverse algorithmic paradigms under identical conditions on a shared target, with experimental validation, remains scarce. Here, we present an integrated AI benchmarking pipeline targeting DHPS and evaluate the full process from computational prediction to experimental validation. We compared 13 MM-GBSA score estimation models and 10 de novo molecular generation methods under 5-seed repeated experiments. For the prediction task, we propose an Out-of-Fold stacking ensemble integrating gradient boosting, graph neural networks, and chemical language models, achieving the highest Spearman rank correlation of 0.861 among all compared models. For the generation task, we propose RL-Design, a reinforcement learning-based framework using the ensemble predictor as a scoring oracle. RL-Design achieved the highest hit rate of 92.02% and maintained drug-like physicochemical properties across generated molecules, outperforming all nine competing methods. Moreover, experimental dose-response assays confirmed inhibitory activity in all three predicted-active compounds, with two highly active compounds showing IC₅₀ values of 31.2 and 49.1 µM. Predicted and experimental activity showed complete concordance across all six tested compounds, providing oracle-independent validation of the pipeline. This pipeline provides a reproducible framework for AI-guided, target-specific inhibitor design and demonstrates the feasibility of HDT-focused antiviral drug discovery through systematic AI benchmarking.Scientific contributionThis work proposes two methods for inhibitor discovery against DHPS, a host-directed antiviral target. A predictive model for MM-GBSA binding-score estimation and a generative model that designs candidate hit compounds both ranked first among previous prediction and generation models under identical benchmarking. The predicted activities were experimentally confirmed by dose-response enzymatic assays.
SN - 1758-2946
ER -