QCMS continues its Online Seminar Series with a new session to be held on May 28, 2026, from 4:00 to 6:15 pm (CET). Thanks to the efforts of the QCMS Webinar Organisation Team, this session brings together an outstanding lineup of invited speakers.
Invited Lecture 1
Entering the Era of Physical AI in Drug Discovery
WOODY SHERMAN
4:00-4:45 pm CET
Abstract: Artificial intelligence is changing how scientific work is performed, but the core challenges of drug discovery remain. Progress still depends on solving difficult predictive problems rooted in the physical world, including molecular interactions, conformational dynamics, binding energetics, chemical reactivity, and the multiparameter tradeoffs that determine whether a molecule can become a medicine. These grand challenges are unlikely to be solved by language-based AI alone. They will require advances in what may be called Physical AI: predictive systems that integrate machine learning with physics, simulation, experiment, and expert judgment. This talk will examine the emerging role of Physical AI in drug discovery and the conditions under which it can materially improve outcomes. It will outline a spectrum of problem types, ranging from those that may benefit primarily from large-scale data-driven learning to those that still require explicit physical modeling or tightly coupled hybrid approaches. It will also discuss the importance of closed-loop workflows that combine AI, simulation, and fit-for-purpose assays, as well as the implications for foundation models, open scientific infrastructure, and the measurable bottlenecks that must be addressed if AI is to improve candidate.
Biography: Woody Sherman is the Founder and Chief Innovation Officer of PsiThera, where he leads the development of a computational-first drug discovery platform advancing oral small-molecule medicines for high-value immunology and inflammation targets traditionally addressed by biologics. He is also Chair of the OpenFold Consortium Executive Committee, guiding a global open science effort to build next-generation foundation models for biomolecular structure and drug design that are shaping the future of the field. Across industry and academia, Woody has been a pioneer in applying physics-based simulation, AI/ML, and integrated experimental-computational workflows to solve real-world drug discovery problems. His career spans leadership roles as Chief Computational Scientist at Roivant Sciences, Chief Scientific Officer at Silicon Therapeutics, and Global Head of Applications Science at Schrödinger, where he helped translate advanced modeling technologies into widely adopted discovery tools. Woody has authored over 100 peer-reviewed publications spanning molecular simulation, free energy methods, protein structure and dynamics, machine learning, and structure-based drug design, and is an Adjunct Professor at the University of Massachusetts Amherst. His work bridges deep technical innovation with strategic vision for how computational technologies can materially transform the speed, cost, and probability of success in drug discovery.
Invited Lecture 2
Deep Learning for Structure-Based Drug Discovery: From Scoring to Generative Design
DAVID R. KOES
4:45-5:30 pm CET
Abstract: Structure-based drug discovery has been transformed by the integration of deep learning, enabling more accurate modeling of protein-ligand interactions and the scalable exploration of chemical space. In this talk, I will present our work developing and applying deep convolutional neural networks (CNNs) for protein-ligand scoring, docking, and virtual screening, with a focus on our open-source docking software GNINA. These models have demonstrated strong performance in both retrospective benchmarks and prospective applications, including results from the CACHE community wide assessment. We additionally describe SPRINT, a vector-based approach for rapidly screening large chemical libraries. We show how SPRINT can be productively incorporated into a deep docking pipeline for virtual screening. He will then discuss how these CNN architectures form the foundation for LiGAN, an early generative model that learns to propose 3D ligand structures directly within a protein binding site. Extending beyond CNNs, I will describe more recent efforts using graph-based deep generative models for both unconditional molecule generation and conditional design with a focus on our state-of-the-art FlowMol flow matching model and recently released OMTRA multi-task generative model.
Biography: David R. Koes is an Associate Professor at the University of Pittsburgh in the Department of Computational and Systems Biology, where he also serves as Vice Chair of Education and Co-Director of the joint Carnegie Mellon–University of Pittsburgh PhD Program in Computational Biology. He received his PhD in Computer Science from Carnegie Mellon University and completed postdoctoral training in computational biology. His research focuses on advancing computational drug discovery through the integration of machine learning, structural biology, and algorithm development. He is a leading contributor to widely used open-source tools such as GNINA and Pharmit, which enable structure-based virtual screening and molecular docking with deep learning. His work has significantly contributed to the application of 3D convolutional neural networks and generative models in drug design, as well as scalable methods for exploring large chemical spaces. Koes is actively involved in the scientific community through peer review, conference organization, and NIH study sections. He has mentored numerous graduate and undergraduate students and contributed extensively to education in computational biology. His research has been recognized with several awards, and his software tools are widely adopted in both academia and industry, supporting real-world drug discovery efforts.
Invited Lecture 3
Boltz: Towards Accurate Biomolecular Modeling and Design
GABRIELE CORSO
5:30-6:15 pm CET
Abstract: Accurately modeling biomolecular interactions is a central challenge in modern biology. Recent advances have substantially improved our ability to predict biomolecular complex structures, understand the strength of interactions and design novel binders. We will present some of our open-source (Boltz-1, Boltz-2 and BoltzGen) and proprietary work pushing the frontier across all of these tasks and how these come together in accurate pipelines for structure-based biologics and small-molecule design.
Biography: Gabriele Corso is the CEO and co-founder of Boltz PBC, a company aiming to advance biomolecular modeling and making it accessible to scientists. Gabriele received his PhD from MIT CSAIL where his research focused on developing novel ML frameworks to tackle challenging problems in drug discovery and he led the development of popular models in the space including DiffDock, Boltz-1 and Boltz-2.
The seminar will be held online, and the access link will be shared only with registered participants.
Please register here:
