Kenneth Atz
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Kenneth Atz has a total of 153 co-authors. Here are the top co-authors by number of shared publications:
- Gisbert Schneider
ETH Zurich
19 shared publications - Uwe Grether
Roche Pharma Research & Early Development, Roche Innovation Center Basel, F. Hoffmann-La Roche Ltd, 4070 Basel, Switzerland
16 shared publications - Rainer E. Martin
Roche Pharma Research and Early Development (pRED), Roche Innovation Center Basel, F. Hoffmann-La Roche Ltd., Basel, Switzerland
10 shared publications - Alex T. Müller
Roche Pharma Research and Early Development (pRED), Roche Innovation Center Basel, F. Hoffmann-La Roche Ltd., Basel, Switzerland
9 shared publications - David F. Nippa
Roche Pharma Research and Early Development (pRED), Roche Innovation Center Basel, F. Hoffmann-La Roche Ltd., Basel, Switzerland
8 shared publications - Clemens Isert
ETH Zurich, Department of Chemistry and Applied Biosciences, Zurich, Switzerland
5 shared publications
Latest Publications
- Open Access
- 9 October 2026
- Published by Springer Nature in Nature Computational Science
- p. 1-17
Advancing chemical reaction prediction in data-scarce drug discovery with active and geometric deep learning
The success of generative molecular design depends on the synthetic accessibility of the molecules it creates. While computational models can help predict chemical reactions, their current effectiveness is limited by a lack of reference data and the difficulty of accurately predicting reaction regioselectivity. Here, to address this, we present an integrated closed-loop workflow that combines efficient reaction data acquisition with high-accuracy deep learning. We developed an active learning strategy, driven by a tree-based ensemble, to guide the acquisition of the most informative laboratory experiments. This experimental campaign yielded a diverse benchmark for C–H borylation. The bespoke dataset allowed us to train and evaluate geometric graph neural networks. We demonstrate that augmenting these symmetry-aware models with self-supervised auxiliary tasks consistently improves performance for both reaction outcome prediction and regioselectivity prediction. Prospective tests on unseen substrates, featuring challenging N -heteroaryl motifs, pinpointed the correct borylation positions in all cases, illustrating the practical utility of our workflow for efficient drug optimization.
- 9 October 2026
- Published by Springer Nature in Nature Computational Science
- p. 1-17
Open AccessAdvancing chemical reaction prediction in data-scarce drug discovery with active and geometric deep learning
The success of generative molecular design depends on the synthetic accessibility of the molecules it creates. While computational models can help predict chemical reactions, their current effectiveness is limited by a lack of reference data and the difficulty of accurately predicting reaction regioselectivity. Here, to address this, we present an integrated closed-loop workflow that combines efficient reaction data acquisition with high-accuracy deep learning. We developed an active learning strategy, driven by a tree-based ensemble, to guide the acquisition of the most informative laboratory experiments. This experimental campaign yielded a diverse benchmark for C–H borylation. The bespoke dataset allowed us to train and evaluate geometric graph neural networks. We demonstrate that augmenting these symmetry-aware models with self-supervised auxiliary tasks consistently improves performance for both reaction outcome prediction and regioselectivity prediction. Prospective tests on unseen substrates, featuring challenging N -heteroaryl motifs, pinpointed the correct borylation positions in all cases, illustrating the practical utility of our workflow for efficient drug optimization.
- Open Access
- 21 August 2026
- Published by American Chemical Society (ACS) in ChemRxiv
Learning from Nature: De novo Generative Design of Natural-Product-Inspired Opioid Agonists
Developing safe and potent analgesics remains a critical global public health priority. Here, we show that generative de novo design can rapidly deliver novel opioid receptor agonists with high synthetic accessibility and desirable pharmacology. Inspired by the natural product mitragynine pseudoindoxyl, a complex metabolite found in traditional medicinal Kratom leaves, we leveraged its µ-opioid receptor binding characteristics to guide our generative computational framework. This deep learning approach successfully translated the partial agonism of the natural product and its G protein signaling bias into novel, structurally distinct chemotypes. One-third of the synthesized compounds exhibited potent receptor activation. Without requiring optimization, lead designs showed favorable pharmacokinetics, minimal off-target activity, and no β-arrestin recruitment. Cryogenic electron microscopy confirmed orthosteric receptor engagement, revealing that sub-pocket interactions alone do not account for their signaling behavior. By shifting away from stochastic high-throughput screening, this nature-inspired, artificial intelligence-driven workflow compresses discovery timelines while maximizing hit rates and molecular quality to deliver next-generation therapeutics.
- 21 August 2026
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessLearning from Nature: De novo Generative Design of Natural-Product-Inspired Opioid Agonists
Developing safe and potent analgesics remains a critical global public health priority. Here, we show that generative de novo design can rapidly deliver novel opioid receptor agonists with high synthetic accessibility and desirable pharmacology. Inspired by the natural product mitragynine pseudoindoxyl, a complex metabolite found in traditional medicinal Kratom leaves, we leveraged its µ-opioid receptor binding characteristics to guide our generative computational framework. This deep learning approach successfully translated the partial agonism of the natural product and its G protein signaling bias into novel, structurally distinct chemotypes. One-third of the synthesized compounds exhibited potent receptor activation. Without requiring optimization, lead designs showed favorable pharmacokinetics, minimal off-target activity, and no β-arrestin recruitment. Cryogenic electron microscopy confirmed orthosteric receptor engagement, revealing that sub-pocket interactions alone do not account for their signaling behavior. By shifting away from stochastic high-throughput screening, this nature-inspired, artificial intelligence-driven workflow compresses discovery timelines while maximizing hit rates and molecular quality to deliver next-generation therapeutics.
- Open Access
- 4 June 2026
- Published by American Chemical Society (ACS) in ChemRxiv
Quantum-accurate strain energy estimation of saturated ring systems in drug-like molecules via equivariant deep learning
Accurate estimation of conformational strain in small-molecule ligands is critical for successful structure-based drug design. While torsional strain in acyclic bonds is readily assessed, evaluating strain in saturated ring systems currently necessitates conformational sampling and compute-intensive quantum-mechanical calculations. To address this limitation, RingAnalyzer is introduced, i.e., a machine learning framework predicting conformational strain directly from single three-dimensional molecular geometries. The method introduces a ring-centric 3D equivariant neural network trained on a rigorously curated dataset of 4’226 drug-like molecules and 27’581 conformations. Calculated at the high-fidelity ωB97XD/Def2TZVP level of theory, this dataset captures diverse stereoelectronic effects. Following extensive data augmentation strategies yielding 76’636 conformations, the computational model accurately predicts strain energies with a mean absolute error of 0.54 kcal/mol, demonstrating robustness to structural artifacts and incomplete structures. Prospective application of RingAnalyzer across diverse structural repositories, including small-molecule conformer generators, 3D generative models, co-folding architectures, docking datasets, the Protein Data Bank, and the Cambridge Structural Database, highlights a widespread occurrence of high-energy ring artifacts. As a computationally efficient and quantum-accurate geometric filter, RingAnalyzer enables the rapid identification and elimination of strained ring systems to streamline structure-based molecular design.
- 4 June 2026
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessQuantum-accurate strain energy estimation of saturated ring systems in drug-like molecules via equivariant deep learning
Accurate estimation of conformational strain in small-molecule ligands is critical for successful structure-based drug design. While torsional strain in acyclic bonds is readily assessed, evaluating strain in saturated ring systems currently necessitates conformational sampling and compute-intensive quantum-mechanical calculations. To address this limitation, RingAnalyzer is introduced, i.e., a machine learning framework predicting conformational strain directly from single three-dimensional molecular geometries. The method introduces a ring-centric 3D equivariant neural network trained on a rigorously curated dataset of 4’226 drug-like molecules and 27’581 conformations. Calculated at the high-fidelity ωB97XD/Def2TZVP level of theory, this dataset captures diverse stereoelectronic effects. Following extensive data augmentation strategies yielding 76’636 conformations, the computational model accurately predicts strain energies with a mean absolute error of 0.54 kcal/mol, demonstrating robustness to structural artifacts and incomplete structures. Prospective application of RingAnalyzer across diverse structural repositories, including small-molecule conformer generators, 3D generative models, co-folding architectures, docking datasets, the Protein Data Bank, and the Cambridge Structural Database, highlights a widespread occurrence of high-energy ring artifacts. As a computationally efficient and quantum-accurate geometric filter, RingAnalyzer enables the rapid identification and elimination of strained ring systems to streamline structure-based molecular design.
- Open Access
- 18 March 2026
- Published by Royal Society of Chemistry (RSC) in Chemical Science
- Vol. 17 (18) , 8976-8988
Single-position ligand modifications tune CB 2 R activity by targeting the toggle switch
This study, exemplified at CB 2 R, showcases how strategic reutilization of a single ligand scaffold fine-tunes activity of a class A GPCR by virtue of targeting its toggle switch.
- 18 March 2026
- Published by Royal Society of Chemistry (RSC) in Chemical Science
- Vol. 17 (18) , 8976-8988
Open AccessSingle-position ligand modifications tune CB 2 R activity by targeting the toggle switch
This study, exemplified at CB 2 R, showcases how strategic reutilization of a single ligand scaffold fine-tunes activity of a class A GPCR by virtue of targeting its toggle switch.
- Open Access
- 26 November 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1)
Expediting hit-to-lead progression in drug discovery through reaction prediction and multi-dimensional optimization
The rapid and economical synthesis of novel bioactive compounds remains a hurdle in drug discovery efforts. This study demonstrates an integrated medicinal chemistry workflow that effectively diversifies hit and lead structures, enabling an acceleration of the critical hit-to-lead optimization phase. Employing high-throughput experimentation (HTE), we generated a comprehensive data set encompassing 13,490 novel Minisci-type C-H alkylation reactions. These data served as the foundation for training deep graph neural networks to accurately predict reaction outcomes. Scaffold-based enumeration of potential Minisci reaction products, starting from moderate inhibitors of monoacylglycerol lipase (MAGL), yielded a virtual library containing 26,375 molecules. This virtual chemical library was evaluated using reaction prediction, physicochemical property assessment, and structure-based scoring, identifying 212 MAGL inhibitor candidates. Of these, 14 compounds were synthesized and exhibited subnanomolar activity, representing a potency improvement of up to 4500 times over the original hit compound. These ligands also showed favorable pharmacological profiles. Co-crystallization of three computationally designed ligands with the MAGL protein provided structural insights into their binding modes. This study demonstrates the potential of combining miniaturized HTE with deep learning and optimization of molecular properties to reduce cycle times in hit-to-lead progression.
- 26 November 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1)
Open AccessExpediting hit-to-lead progression in drug discovery through reaction prediction and multi-dimensional optimization
The rapid and economical synthesis of novel bioactive compounds remains a hurdle in drug discovery efforts. This study demonstrates an integrated medicinal chemistry workflow that effectively diversifies hit and lead structures, enabling an acceleration of the critical hit-to-lead optimization phase. Employing high-throughput experimentation (HTE), we generated a comprehensive data set encompassing 13,490 novel Minisci-type C-H alkylation reactions. These data served as the foundation for training deep graph neural networks to accurately predict reaction outcomes. Scaffold-based enumeration of potential Minisci reaction products, starting from moderate inhibitors of monoacylglycerol lipase (MAGL), yielded a virtual library containing 26,375 molecules. This virtual chemical library was evaluated using reaction prediction, physicochemical property assessment, and structure-based scoring, identifying 212 MAGL inhibitor candidates. Of these, 14 compounds were synthesized and exhibited subnanomolar activity, representing a potency improvement of up to 4500 times over the original hit compound. These ligands also showed favorable pharmacological profiles. Co-crystallization of three computationally designed ligands with the MAGL protein provided structural insights into their binding modes. This study demonstrates the potential of combining miniaturized HTE with deep learning and optimization of molecular properties to reduce cycle times in hit-to-lead progression.
- 15 October 2025
- Published by Royal Society of Chemistry (RSC) in RSC Medicinal Chemistry
- Vol. 16 (12) , 5753-5764
Accelerating compound synthesis in drug discovery: the role of digitalisation and automation
This opinion addresses how digitalisation and automation can reduce the synthesis bottleneck in the DMTA cycle. Current automated synthesis & planning, future data-rich integrated platforms, and the medicinal chemist's evolving role are reviewed.
- 15 October 2025
- Published by Royal Society of Chemistry (RSC) in RSC Medicinal Chemistry
- Vol. 16 (12) , 5753-5764
Accelerating compound synthesis in drug discovery: the role of digitalisation and automation
This opinion addresses how digitalisation and automation can reduce the synthesis bottleneck in the DMTA cycle. Current automated synthesis & planning, future data-rich integrated platforms, and the medicinal chemist's evolving role are reviewed.
- Open Access
- 14 July 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Active and geometric deep learning advances chemical reaction prediction in data-scarce drug discovery
Late-stage functionalization (LSF) is a critical approach for optimizing complex molecules during the advanced stages of drug discovery. Although computational models offer a pivotal opportunity to enhance LSF by predicting reaction outcome, their current impact is constrained by insufficient reference data and the challenge of accurately predicting reaction regioselectivity. To address these critical limitations, we present an integrated closed-loop workflow that combines efficient reaction data acquisition with high-accuracy deep learning. We developed an active learning strategy, driven by a computationally efficient tree-based ensemble, to guide the acquisition of the most informative laboratory experiments. This experimental campaign produced new reactions on novel and complex substrates, yielding the most diverse benchmark yet reported for C–H borylation. The profoundly expanded bespoke dataset allowed us to train and evaluate geometric graph neural networks. We demonstrate that augmenting these symmetry-aware models with self-supervised auxiliary tasks consistently improves performance for both reaction outcome prediction and regioselectivity prediction. Prospective tests on unseen substrates, featuring challenging N-heteroaryl motifs, pinpointed the correct borylation positions in all cases, thereby closing the loop between model-guided experimentation and deep learning, and illustrating the practical utility of our workflow for efficient drug optimization.
- 14 July 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessActive and geometric deep learning advances chemical reaction prediction in data-scarce drug discovery
Late-stage functionalization (LSF) is a critical approach for optimizing complex molecules during the advanced stages of drug discovery. Although computational models offer a pivotal opportunity to enhance LSF by predicting reaction outcome, their current impact is constrained by insufficient reference data and the challenge of accurately predicting reaction regioselectivity. To address these critical limitations, we present an integrated closed-loop workflow that combines efficient reaction data acquisition with high-accuracy deep learning. We developed an active learning strategy, driven by a computationally efficient tree-based ensemble, to guide the acquisition of the most informative laboratory experiments. This experimental campaign produced new reactions on novel and complex substrates, yielding the most diverse benchmark yet reported for C–H borylation. The profoundly expanded bespoke dataset allowed us to train and evaluate geometric graph neural networks. We demonstrate that augmenting these symmetry-aware models with self-supervised auxiliary tasks consistently improves performance for both reaction outcome prediction and regioselectivity prediction. Prospective tests on unseen substrates, featuring challenging N-heteroaryl motifs, pinpointed the correct borylation positions in all cases, thereby closing the loop between model-guided experimentation and deep learning, and illustrating the practical utility of our workflow for efficient drug optimization.
- Open Access
- 7 February 2025
- Published by Wiley in Chemistry – A European Journal
- Vol. 31 (24) , e202404517
Discovery of 1,3,4‐Oxadiazin‐5‐One Derivative CJ1‐34 as a Partial ATP Synthase Inhibitor for CNS Applications
ATP synthase dysregulation has been implicated in many diseases, including cancer and neurodegenerative diseases. Whilst ATP synthase‐targeting compounds have been reported, most are large or polar compounds and lack appropriate properties for a CNS drug. We designed, synthesised, and evaluated a novel series of ATP synthase targeting compounds, resulting in a 1,3,4‐oxadiazin‐5‐one scaffold with improved physiochemical properties. In vitro evaluation of our library led to the discovery of CJ1‐34 as a partial ATP synthase inhibitor with a determined IC 50 of 394 nM in isolated bovine mitochondria. Molecular docking experiments demonstrated that calculated docking scores aligned with the observed in vitro pharmacology and supports the F 1 region of the ATP synthase as the potential binding site. Overall, this provides the foundation to further investigate CJ1‐34 as a potential CNS drug for diseases where mitochondria are implicated, including Alzheimer's and Parkinson's disease.
- 7 February 2025
- Published by Wiley in Chemistry – A European Journal
- Vol. 31 (24) , e202404517
Open AccessDiscovery of 1,3,4‐Oxadiazin‐5‐One Derivative CJ1‐34 as a Partial ATP Synthase Inhibitor for CNS Applications
ATP synthase dysregulation has been implicated in many diseases, including cancer and neurodegenerative diseases. Whilst ATP synthase‐targeting compounds have been reported, most are large or polar compounds and lack appropriate properties for a CNS drug. We designed, synthesised, and evaluated a novel series of ATP synthase targeting compounds, resulting in a 1,3,4‐oxadiazin‐5‐one scaffold with improved physiochemical properties. In vitro evaluation of our library led to the discovery of CJ1‐34 as a partial ATP synthase inhibitor with a determined IC 50 of 394 nM in isolated bovine mitochondria. Molecular docking experiments demonstrated that calculated docking scores aligned with the observed in vitro pharmacology and supports the F 1 region of the ATP synthase as the potential binding site. Overall, this provides the foundation to further investigate CJ1‐34 as a potential CNS drug for diseases where mitochondria are implicated, including Alzheimer's and Parkinson's disease.
- Open Access
- 28 January 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Expediting hit-to-lead progression in drug discovery through reaction prediction and multi-objective molecular optimization
The rapid and economical synthesis of novel bioactive compounds remains a significant hurdle in drug discovery efforts. This study demonstrates an integrated medicinal chemistry workflow that effectively diversifies hit and lead structures, enabling an efficient acceleration of the critical hit-to-lead optimization phase. Employing high-throughput experimentation (HTE), we generated a comprehensive data set encompassing 13,490 novel Minisci-type C-H alkylation reactions. This data set served as the foundation for training deep graph neural networks to accurately predict reaction outcomes. Scaffold-based enumeration of potential Minisci reaction products, starting from moderate inhibitors of monoacylglycerol lipase (MAGL), yielded a virtual library containing 26,375 molecules. This virtual chemical library was evaluated using reaction prediction, physicochemical property assessment, and structure-based scoring, identifying 212 potential MAGL inhibitor candidates. Of these, 14 ligands were synthesized and exhibited subnanomolar activity, representing a potency improvement of up to 4500 times over the original hit compound. These compounds also displayed favorable pharmacological profiles. Co-crystallization of three computationally designed ligands with the MAGL protein provided valuable structural insights into their preferred binding poses. This study demonstrates the potential of combining miniaturized HTE with deep learning and molecular property optimization to reduce cycle times in drug discovery.
- 28 January 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessExpediting hit-to-lead progression in drug discovery through reaction prediction and multi-objective molecular optimization
The rapid and economical synthesis of novel bioactive compounds remains a significant hurdle in drug discovery efforts. This study demonstrates an integrated medicinal chemistry workflow that effectively diversifies hit and lead structures, enabling an efficient acceleration of the critical hit-to-lead optimization phase. Employing high-throughput experimentation (HTE), we generated a comprehensive data set encompassing 13,490 novel Minisci-type C-H alkylation reactions. This data set served as the foundation for training deep graph neural networks to accurately predict reaction outcomes. Scaffold-based enumeration of potential Minisci reaction products, starting from moderate inhibitors of monoacylglycerol lipase (MAGL), yielded a virtual library containing 26,375 molecules. This virtual chemical library was evaluated using reaction prediction, physicochemical property assessment, and structure-based scoring, identifying 212 potential MAGL inhibitor candidates. Of these, 14 ligands were synthesized and exhibited subnanomolar activity, representing a potency improvement of up to 4500 times over the original hit compound. These compounds also displayed favorable pharmacological profiles. Co-crystallization of three computationally designed ligands with the MAGL protein provided valuable structural insights into their preferred binding poses. This study demonstrates the potential of combining miniaturized HTE with deep learning and molecular property optimization to reduce cycle times in drug discovery.
- Open Access
- 23 January 2025
- Published by Wiley in Molecular Informatics
- Vol. 44 (1) , e202400361
Simple User‐Friendly Reaction Format
Utilizing the growing wealth of chemical reaction data can boost synthesis planning and increase success rates. Yet, the effectiveness of machine learning tools for retrosynthesis planning and forward reaction prediction relies on accessible, well‐curated data presented in a structured format. Although some public and licensed reaction databases exist, they often lack essential information about reaction conditions. To address this issue and promote the principles of findable, accessible, interoperable, and reusable (FAIR) data reporting and sharing, we introduce the Simple User‐Friendly Reaction Format (SURF). SURF standardizes the documentation of reaction data through a structured tabular format, requiring only a basic understanding of spreadsheets. This format enables chemists to record the synthesis of molecules in a format that is understandable by both humans and machines, which facilitates seamless sharing and integration directly into machine learning pipelines. SURF files are designed to be interoperable, easily imported into relational databases, and convertible into other formats. This complements existing initiatives like the Open Reaction Database (ORD) and Unified Data Model (UDM). At Roche, SURF plays a crucial role in democratizing FAIR reaction data sharing and expediting the chemical synthesis process.
- 23 January 2025
- Published by Wiley in Molecular Informatics
- Vol. 44 (1) , e202400361
Open AccessSimple User‐Friendly Reaction Format
Utilizing the growing wealth of chemical reaction data can boost synthesis planning and increase success rates. Yet, the effectiveness of machine learning tools for retrosynthesis planning and forward reaction prediction relies on accessible, well‐curated data presented in a structured format. Although some public and licensed reaction databases exist, they often lack essential information about reaction conditions. To address this issue and promote the principles of findable, accessible, interoperable, and reusable (FAIR) data reporting and sharing, we introduce the Simple User‐Friendly Reaction Format (SURF). SURF standardizes the documentation of reaction data through a structured tabular format, requiring only a basic understanding of spreadsheets. This format enables chemists to record the synthesis of molecules in a format that is understandable by both humans and machines, which facilitates seamless sharing and integration directly into machine learning pipelines. SURF files are designed to be interoperable, easily imported into relational databases, and convertible into other formats. This complements existing initiatives like the Open Reaction Database (ORD) and Unified Data Model (UDM). At Roche, SURF plays a crucial role in democratizing FAIR reaction data sharing and expediting the chemical synthesis process.