QCMS is launching an Online Seminar Series. The inaugural session will take place on Tuesday, February 10, 2026, from 4:00 to 5:30 pm (CET).
The seminar will be held via Microsoft Teams, and the access link will be shared only with registered participants. If you are planning to attend please register here.
Further details of the lecture are provided below.
Invited Lecture 1
Reminiscing about the Future of Cheminformatics.
Alexander Tropsha
UNC Eshelman School of Pharmacy, UNC-Chapel Hill, Chapel Hill, NC, USA.
February 10, 2026 (Tuesday) 4:00pm CET.
Abstract: The field of Cheminformatics has always been one of the earliest adopters of innovations in computational data-analytical methods. Multiple algorithms leveraging fundamental advances in ML such as (deep) neural networks, multi-dimensional scaling, generative topographic mapping, support vector machines, natural language processing, generative AI and other approaches have contributed to the evolution of the field. However, one may argue that its major challenges have remained unchanged including tasks such as chemical similarity searching, QSAR modeling, molecular docking, data visualization, and rational design of new chemical entities predicted to have the desired property or activity. Thus, the history of the field can provide hints about its future, and I will review how computational tools that address fundamental cheminformatics challenges have evolved with the major transformative component of the field: the continuing growth of biomolecular datasets including recent Big Bang expansion of synthetically feasible and purchasable Chemical Universe. These advances have created substantial computational challenges for traditional approaches to virtual screening (VS) such as similarity searching and molecular docking used in early phase drug discovery. I will describe novel resource- and cost-effective approaches to both ligand-based (LB) and structure-based (SB) VS with the focus on experimentally testable hypotheses generation. I will also discuss emerging methods for knowledge mining across multiple databases integrated into a biomedical knowledge graph to support target discovery and drug repurposing. In place of using LLM models, I will end with an attempt to hallucinate about the future of cheminformatics including greater integration of computations with experiment in self-driving labs, proliferation of cheminformatics tools and concepts across multiple disciplines, and democratization of drug discovery as part of imminent exponential growth of our field.
Speaker Profile:
Alexander Tropsha, PhD is K.H. Lee Distinguished Professor at the UNC Eshelman School of Pharmacy, and Chief Domain Scientist for Molecular Informatics at the Renaissance Computing Institute (RENCI), UNC-Chapel Hill. Prof. Tropsha obtained his PhD in Chemical Enzymology in 1986 from Moscow State University, Russia. His main research interests are in the area of Data Science with applications to drug discovery, chemical safety predictions including computational and combinatorial NAMs methods, and materials science. He has authored or co-authored ca. 350 peer-reviewed research papers, reviews, and book chapters. He is an elected Fellow of the American Institute for Medical and Biological Engineering (AIMBE) and a consultant to several technology and drug discovery companies.
Invited Lecture 2
Rethinking Bioactivity Modeling: Representation, Context, and Predictive Space
Karina Martinez Mayorga
Instituto de Quimica, UNAM, Mexico
February 10, 2026 (Tuesday) 4:45pm CET.
Abstract: Predictive models in cheminformatics are commonly built under the assumption that bioactivity varies smoothly with molecular structure, yet recent work shows that changes in experimental context, such as dose range and data distribution, can lead to abrupt losses of predictivity even when the underlying chemical space remains unchanged. This talk discusses the distinction between chemical space and predictive space, emphasizing that model performance depends on how molecular information is represented and selected rather than on structural similarity alone, and presents descriptor selection as an active step in constructing chemical space, with brief reference to alternative representations, including quantum-derived descriptors. In parallel, recent discussions on Structure–Property Associations are introduced as a complementary perspective that questions the notion of “association” itself when linking structure to biological properties, while intrinsically multidimensional response systems, such as biased agonism, are mentioned as conceptual examples highlighting broader challenges for predictive modeling in complex biological settings.
Speaker Profile:
Karina Martínez-Mayorga, PhD is a researcher in cheminformatics and computational chemical biology at the Institute of Chemistry, National Autonomous University of Mexico (UNAM). Her research focuses on predictive modeling of bioactivity, representation, and descriptor selection, with particular attention to the limitations of QSAR and machine-learning models in complex biological systems. She has published extensively in leading journals in the field and currently serves as Editor-in-Chief of the Journal of Cheminformatics.
