IMPECCABLE: A Digital Twin Drug Discovery Workflow Comprising AI, Physics and Experiment & the Creation of Nanomolar-Potent Inhibitors
17 affiliations
- 1Argonne Leadership Computing Facility
- 2Argonne National Laboratory
- 3Wadhwani School of Data Science and Artificial Intelligence
- 4Indian Institute of Technology Madras
- 5Department of Chemistry
- 6University College London
- 7Molecular AI,
- 8R&D, AstraZeneca
- 9School of Pharmacy, University College London, 29-39 Brunswick Square, London, United Kingdom, WC1N 1AX
- 10Structural Genomics Consortium
- 11University Health Network
- 12Department of Pharmacology and Toxicology
- 13University of Toronto
- 14Princess Margaret Cancer Centre
- 15Oak Ridge Leadership Computing Facility, Oak Ridge National Laboratory
- 16Institute for Data Science in Oncology
- 17UT MD Anderson Cancer Center
Abstract
The traditional drug discovery process is slow, costly and prone to high failure rates. To address this, we have developed IMPECCABLE, an integrated, heterogeneous in silico workflow combining physics-based methods and artificial intelligence to explore small-molecule chemical space. Operating on Frontier and Aurora exascale supercomputers, the workflow rapidly screens, generates and evaluates candidate compounds. Promising candidates enter a generative, multi-objective active learning loop where reliable binding free energies from ESMACS and TIES guide REINVENT to predict diverse, novel and synthesizable molecules. Unlike conventional approaches, this tightly coupled architecture integrates hit identification together with hit-to-lead optimisation and expansion within a single automated pipeline. Crucially, and uniquely, IMPECCABLE also integrates human medicinal chemistry input based on experiments, creating a molecular digital twin. We applied IMPECCABLE to two disease-related targets, WDR91 and TNKS2, demonstrating its ability to produce viable binders, while also performing prospective experimental validation for WDR91. Following multi-objective optimisation via an iterative REINVENT-TIES cycle, we identified four potent hits for WDR91 exhibiting nanomolar binding affinity with KD values between 129 ± 1 nM and 287 ± 1 nM. Remarkably, these four hits emerged from a selection of just ten synthesized compounds, demonstrating an exceptional ca 40% success rate.
Related
- Discovery of Nanomolar CDK2 Inhibitors Using Boltz-2
- Discovery of selective, metabolically stable pyrazole-based FLT3 inhibitors for the treatment of acute myeloid leukemia
- Machine-learning-guided discovery of selective CK2 inhibitors with reduced CDK off-target activity
- Constructive chemical discovery through autonomous reasoning and experiment
- Fragment-Based Discovery of KLK6 and KLK7 Inhibitors
BibTeX
@misc{jiang2026impeccable,
title = {{IMPECCABLE: A Digital Twin Drug Discovery Workflow Comprising AI, Physics and Experiment & the Creation of Nanomolar-Potent Inhibitors}},
author = {Wei Jiang and Agastya P Bhati and Shunzhou Wan and Hannes H Loeffler and Marco Klähn and Mateusz Bieniek and Kangping Liu and Shabbir Ahmad and Levon Halabelian and Jinyu Dong and Xibei Zhang and Matthieu Schapira and Conrad Veranso Simoben and Justin Wozniak and Ketan Maheshwari and Eric Stahlberg and Darren V Green and Matthew Todd and Peter V Coveney},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15002535/v3},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15002535/v3},
note = {CC BY},
}
RIS
TY - UNPB
TI - IMPECCABLE: A Digital Twin Drug Discovery Workflow Comprising AI, Physics and Experiment & the Creation of Nanomolar-Potent Inhibitors
AU - Wei Jiang
AU - Agastya P Bhati
AU - Shunzhou Wan
AU - Hannes H Loeffler
AU - Marco Klähn
AU - Mateusz Bieniek
AU - Kangping Liu
AU - Shabbir Ahmad
AU - Levon Halabelian
AU - Jinyu Dong
AU - Xibei Zhang
AU - Matthieu Schapira
AU - Conrad Veranso Simoben
AU - Justin Wozniak
AU - Ketan Maheshwari
AU - Eric Stahlberg
AU - Darren V Green
AU - Matthew Todd
AU - Peter V Coveney
PY - 2026
DA - 2026/10/20
PB - ChemRxiv
DO - 10.26434/chemrxiv.15002535/v3
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15002535/v3
AB - The traditional drug discovery process is slow, costly and prone to high failure rates. To address this, we have developed IMPECCABLE, an integrated, heterogeneous in silico workflow combining physics-based methods and artificial intelligence to explore small-molecule chemical space. Operating on Frontier and Aurora exascale supercomputers, the workflow rapidly screens, generates and evaluates candidate compounds. Promising candidates enter a generative, multi-objective active learning loop where reliable binding free energies from ESMACS and TIES guide REINVENT to predict diverse, novel and synthesizable molecules. Unlike conventional approaches, this tightly coupled architecture integrates hit identification together with hit-to-lead optimisation and expansion within a single automated pipeline. Crucially, and uniquely, IMPECCABLE also integrates human medicinal chemistry input based on experiments, creating a molecular digital twin. We applied IMPECCABLE to two disease-related targets, WDR91 and TNKS2, demonstrating its ability to produce viable binders, while also performing prospective experimental validation for WDR91. Following multi-objective optimisation via an iterative REINVENT-TIES cycle, we identified four potent hits for WDR91 exhibiting nanomolar binding affinity with KD values between 129 ± 1 nM and 287 ± 1 nM. Remarkably, these four hits emerged from a selection of just ten synthesized compounds, demonstrating an exceptional ca 40% success rate.
ER -