Robust out-of-distribution prediction of Buchwald–Hartwig reactions
10 affiliations
- 1Drug Discovery Data Science, In Silico Discovery, Johnson & Johnson, Porto Salvo, Portugal
- 2Laboratory of Artificial Chemical Intelligence (LIAC), Institut des Sciences et Ingénierie Chimiques, Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland
- 3Johnson & Johnson (United States)
- 4École Polytechnique Fédérale de Lausanne
- 5Chemistry Capabilities, Analytical and Purification, Global Discovery Chemistry, Johnson & Johnson, Spring House, PA, USA
- 6Drug Discovery Data Science, In Silico Discovery, Johnson & Johnson, Spring House, PA, USA
- 7Drug Discovery Data Science, In Silico Discovery, Johnson & Johnson, Cambridge, MA, USA. [email protected]
- 8Laboratory of Artificial Chemical Intelligence (LIAC), Institut des Sciences et Ingénierie Chimiques, Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland. [email protected]
- 9National Centre of Competence in Research (NCCR) Catalysis, Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland. [email protected]
- 10Chemistry Capabilities, Analytical and Purification, Global Discovery Chemistry, Johnson & Johnson, Spring House, PA, USA. [email protected]
Abstract
The Buchwald-Hartwig cross-coupling is a cornerstone of modern pharmaceutical synthesis, yet predictive modeling of its outcomes remains constrained by data quality and chemical space coverage. Electronic laboratory notebooks contain heterogeneous, noisy records, while open-source high-throughput experimentation (HTE) datasets are fragmented and narrow in scope, leading to poor model performance on unseen substrates and conditions. Here we introduce a framework that systematically standardizes and integrates multiple reaction datasets into a high-quality, unique-structure-per-entity dataset, coupled with active learning to strategically expand chemical space. By merging published Buchwald-Hartwig HTE data with new experimental results, we achieve a model with predictive power across novel substrates and conditions, delivering improved out-of-distribution predictions compared with previous approaches. Crucially, model-guided reagent recommendations were validated experimentally, confirming the framework's utility to uncover unexplored reactivity. This work establishes a blueprint for robust machine learning in synthetic chemistry and enables preemptive in silico reagent screening to accelerate pharmaceutical discovery.
Related
- Relay Catalysis for Selective Mono- N -methylation and Buchwald–Hartwig Amination of Primary Aromatic and Aliphatic Amines Using a Vaska-Type Ir and Ir/Pd Catalysts
- Evaluating a Lightweight Neural Network for Aggregate Isotope Distribution Prediction
- CCSBase2: A Unified Benchmark Dataset and Machine Learning Framework for Robust CCS Prediction.
- Is the prediction of fluoroquinolones adsorption in soils still valid in presence of metal trace elements? - Towards a robust predictive approach
- Excited-State Photophysics Improves Scaffold-Robust Phototoxicity Prediction and Reveals Compound-Specific State Effects
BibTeX
@article{neves2026robust,
title = {{Robust out-of-distribution prediction of Buchwald–Hartwig reactions}},
author = {Paulo Neves and Bo Hao and Santeri Aikonen and Justin B. Diccianni and Jörg K. Wegner and Philippe Schwaller and Iulia I. Strambeanu},
journal = {Nature Computational Science},
year = {2026},
month = aug,
doi = {10.1038/s43588-026-01017-6},
url = {https://doi.org/10.1038/s43588-026-01017-6},
note = {CC BY},
}
RIS
TY - JOUR
TI - Robust out-of-distribution prediction of Buchwald–Hartwig reactions
AU - Paulo Neves
AU - Bo Hao
AU - Santeri Aikonen
AU - Justin B. Diccianni
AU - Jörg K. Wegner
AU - Philippe Schwaller
AU - Iulia I. Strambeanu
PY - 2026
DA - 2026/08/10
JO - Nature Computational Science
DO - 10.1038/s43588-026-01017-6
UR - https://doi.org/10.1038/s43588-026-01017-6
AB - The Buchwald-Hartwig cross-coupling is a cornerstone of modern pharmaceutical synthesis, yet predictive modeling of its outcomes remains constrained by data quality and chemical space coverage. Electronic laboratory notebooks contain heterogeneous, noisy records, while open-source high-throughput experimentation (HTE) datasets are fragmented and narrow in scope, leading to poor model performance on unseen substrates and conditions. Here we introduce a framework that systematically standardizes and integrates multiple reaction datasets into a high-quality, unique-structure-per-entity dataset, coupled with active learning to strategically expand chemical space. By merging published Buchwald-Hartwig HTE data with new experimental results, we achieve a model with predictive power across novel substrates and conditions, delivering improved out-of-distribution predictions compared with previous approaches. Crucially, model-guided reagent recommendations were validated experimentally, confirming the framework's utility to uncover unexplored reactivity. This work establishes a blueprint for robust machine learning in synthetic chemistry and enables preemptive in silico reagent screening to accelerate pharmaceutical discovery.
SN - 2662-8457
ER -