See my most recent publications on Google Scholar.
Published in arXiv, 2024
Molecular generative models can directly optimize for constrained synthesizability.
Recommended citation: J. Guo, P. Schwaller “It Takes Two to Tango: Directly Optimizing for Constrained Synthesizability in Generative Molecular Design”, arXiv, 2024. Paper Link
Published in arXiv, 2024
Molecular generative models can directly optimize for synthesizability using retrosynthesis models.
Recommended citation: J. Guo, P. Schwaller “Directly Optimizing for Synthesizability in Generative Molecular Design using Retrosynthesis Models”, arXiv, 2024. Paper Link
Published in Nature Machine Intelligence, 2024
Review on generative molecular design for small molecule drug discovery.
Recommended citation: Y. Du§, A. R. Jamasb§, J. Guo§, T. Fu, C. Harris, Y. Wang, C. Duan, P. Liò, P. Schwaller, T. L. Blundell, Nat Mach Intell, 2024. Paper Link
Published in arXiv, 2024
Saturn is a framework for sample-efficient molecular generative design with state-of-the-art sample efficiency.
Recommended citation: J. Guo, P. Schwaller “Saturn: Sample-efficient Generative Molecular Design using Memory Manipulation”, arXiv, 2024. Paper Link
Published in JACS Au, 2024
Here, we propose Augmented Memory to combine experience replay with SMILES augmentation to drastically improve sample efficiency in molecular design.
Recommended citation: J. Guo, P. Schwaller “Augmented Memory: Capitalizing on Experience Replay to Accelerate De Novo Molecular Design”, JACS Au, 2024. Paper Link
Published in Chemical Science, 2024
Here, we combine active learning with reinforcement learning to improve sample efficiency in molecular design.
Recommended citation: M. Dodds, J. Guo, T. Löhr, A. Tibo, O. Engkvist, J. P. Janet “Sample Efficient Reinforcement Learning with Active Learning for Molecular Design”, Chem. Sci., 2024. Paper Link
Published in ICLR 2024, 2023
Here, we show that extracting molecular substructures from molecular generative models provides a notion of explainability and can be exploited for sample efficiency gains.
Recommended citation: J. Guo, P. Schwaller “Beam Enumeration: Probabilistic Explainability For Sample Efficient Self-conditioned Molecular Design”, ICLR, 2024. Paper Link
Published in CHIMIA, 2023
Review on Bayesian optimization for chemical reactions.
Recommended citation: J. Guo§, B. Ranković§, P. Schwaller “Bayesian Optimization for Chemical Reactions”, CHIMIA, 2023, 77, 31-38. Paper Link
Published in Digital Discovery, 2023
Link-INVENT is an extension of REINVENT (generative molecular design with reinforcement learning) for the design of chemical linkers between two molecular sub-units. Link-INVENT is suitable for fragment linking, scaffold hopping, and PROTAC linker design and allows explicit control over optimizing linker-specific properties such as length.
Recommended citation: J. Guo§, F. Knuth§, C. Margreitter, J. P. Janet, K. Papadopoulos, O. Engkvist, A. Patronov “Link-INVENT: Generative Linker Design with Reinforcement Learning”, Digital Discovery, 2023. Paper Link
Published in Bioinformatics, 2022
Icolos is a workflow manager to automate computational chemistry calculations. Built-in integration with REINVENT (generative molecular design with reinforcement learning) allows various computational chemistry end-points to be explicitly optimized during the generative process.
Recommended citation: J. H. Moore, M. R. Bauer, J. Guo, A. Patronov, O. Engkvist, C. Margreitter “Icolos: A workflow manager for structure based post-processing of de novo generated small molecules”, Bioinformatics, 2022, 38(21), 4951-4952. Paper Link
Published in Nat. Mach. Intell, 2022
Here, we implement curriculum learning in REINVENT (generative molecular design with reinforcement learning) to decompose complex molecular design objectives into a sequence of simpler intermediate tasks. Compared to standard reinforcement learning, target objective convergence can be greatly accelerated.
Recommended citation: J. Guo§, V. Fialková§, J. D. Arango, C. Margreitter, J. P. Janet, K. Papadopoulos, O. Engkvist, A. Patronov “Improving De Novo Molecular Design with Curriculum Learning”, Nat Mach Intell, 2022, 4, 555–563. Paper Link
Published in J. Cheminform, 2021
DockStream is a software wrapper providing access to multiple molecular docking algorithms. Built-in integration with REINVENT (generative molecular design with reinforcement learning) allows molecular docking scores to be optimized during the generative process.
Recommended citation: J. Guo, J. P. Janet, M. R. Bauer, E. Nittinger, K. A. Giblin, K. Papadopoulos, A. Voronov, A. Patronov, O. Engkvist, C. Margreitter “DockStream: A Docking Wrapper to Enhance De Novo Molecular Design”, J. Cheminform, 2021, 13, 89. Paper Link