Plenary talk 1
CAR T Cell Therapies and Strategies to Enhance Immune Responses Against Solid Tumors
Maria Ormhøj, PhD
Assistant Professor, University of Southern Denmark
Abstract
Chimeric antigen receptor (CAR) T-cell therapy has revolutionized the treatment of several hematological malignancies, yet translating this success to solid tumors remains a major challenge. Unlike blood cancers, solid tumors present multiple barriers to effective therapy, including limited T-cell infiltration, antigen heterogeneity, immunosuppressive tumor microenvironments, and tumor-intrinsic resistance to immune-mediated killing. In this talk, I will discuss the key biological mechanisms that currently limit CAR T-cell efficacy in solid tumors and provide perspectives on how next-generation engineering approaches may help overcome these obstacles. Topics will include strategies to improve trafficking and persistence, address antigen escape, enhance resistance to immunosuppression, and reprogram tumor cell death pathways. I will also highlight emerging concepts that move beyond conventional CAR design, including the use of synthetic biology and engineered protein delivery systems to directly modulate resistance mechanisms within cancer cells. Together, these approaches may help unlock the full potential of cellular immunotherapy for patients with solid tumors.
Biography
Maria Ormhøj, PhD, is Assistant Professor at the University of Southern Denmark, where she leads a research program focused on next-generation CAR T-cell engineering. Her work combines synthetic biology, protein engineering, and cancer immunology to develop novel cellular therapies capable of overcoming resistance mechanisms in cancer. Dr. Ormhøj completed her PhD in Cancer Immunology at the University of Southern Denmark and conducted research at Harvard Medical School, Leibniz Institute for Immunotherapy Regensburg and Technical University of Denmark. Her research focuses on improving CAR T-cell efficacy in both hematological malignancies and solid tumors through innovative approaches that target tumor resistance, antigen heterogeneity, and immune evasion.
Plenary Talk 2
High-Throughput Drug Discovery for Previously Undruggable Receptors: Challenges and Opportunities
Fernando Salgado Polo, PhD
Postdoctoral Fellow, University of Copenhagen
Abstract
G protein–coupled receptors (GPCRs) are among the most important drug targets, accounting for roughly one-third of approved therapeutics. Yet a substantial fraction of this receptor family − particularly chemokine and lipid-binding receptors − remain difficult to target with traditional drug discovery approaches. Technical challenges, including instability of these membrane proteins and limitations in functional screening, have left a significant portion of this receptor family effectively “undruggable.”
In this work, we explored a high-throughput strategy to overcome these barriers using the RaPID (Random Nonstandard Peptides Integrated Discovery) platform. This approach enables the screening of trillion-size libraries of cyclic peptides to identify potent receptor binders. Using minimal amounts of purified receptors stabilized in nanodiscs, we demonstrated efficient screening against GPCRs that have historically been difficult to access.
We first validated the approach on a well-studied receptor (CXCR4), identifying highly potent and selective peptides effectively blocked receptor activation. We then extended the platform to two previously “undruggable” receptors (XCR1 and LPAR6), generating the first selective ligands capable of suppressing their activity and cellular responses relevant to disease.
From an artificial intelligence (AI) perspective, this work highlights both the opportunities and current bottlenecks in high-throughput drug discovery for complex membrane proteins. Machine-learning approaches could exploit the information-rich datasets generated by RaPID (or similar platforms) to connect sequence, structure and potential function, thereby guiding library design and accelerating ligand discovery.
Biography
Fernando Salgado-Polo is a biochemist specializing in drug discovery and cancer research. He earned his PhD in Biochemistry at The Netherlands Cancer Institute (NKI), where he determined the structure–activity relationships of novel compounds targeting the extracellular enzyme autotaxin (ATX). Building on this foundation, he currently carries out his postdoctoral research at the University of Copenhagen (UCPH), where he employs high-throughput approaches to discover cyclic peptide-based drugs against membrane receptors that have so far remained inaccessible to conventional approaches. His research has received support from the Danish Cancer Society, enabling him to further translate his findings into immunotherapy applications to combat cancer.
Plenary Talk 3
Responsible AI and Mathematical Foundations of Machine Unlearning
Vinay Chakravarthi Gogineni, PhD
Associate Professor, Applied AI and Data Science, University of Southern Denmark
Short Abstract
As AI systems become increasingly embedded in our daily lives, concerns surrounding privacy, data ownership, and the right to be forgotten are gaining unprecedented importance. While deleting data from storage is relatively straightforward, ensuring that an AI model truly forgets information it has already learned remains a fundamental challenge.
In this talk, I will introduce the emerging field of machine unlearning, which seeks to remove the influence of specific data from trained AI models without requiring complete retraining. I will discuss the key challenges, recent advances, and practical implications of machine unlearning, particularly in the context of privacy-preserving and trustworthy AI. Additionally, I will share insights from our latest research on developing AI systems that better support user privacy, regulatory compliance, and responsible data governance.
Biography
Vinay Chakravarthi Gogineni is an Associate Professor in the Applied AI and Data Science Unit, where he conducts research aimed at advancing AI systems that are safe, responsible, and trustworthy. He completed his PhD in Decentralized Multitask Learning at the Indian Institute of Technology Kharagpur and has conducted research at the Norwegian University of Science and Technology, Trondheim, Norway, and the Simula Metropolitan Center for Digital Engineering, Oslo, Norway.
Plenary Talk 4
Open-Source Direct Storage (OpenDS) for AI Infrastructure
Nikhil B Gaikwad, PhD
Software Engineer
Samsung Semiconductor Denmark Research
Abstract
OpenDS is an open-source, device-agnostic initiative designed to deliver a high-performance, direct storage access layer for AI workloads across heterogeneous computing platforms. Driven by the need for flexible and interoperable I/O infrastructure in modern AI systems, OpenDS leverages Samsung’s AiSIO technology to provide a unified interface that enables highly efficient data movement directly between storage devices like NVMe and AI accelerator memory, independent of the underlying hardware architecture. The project's methodology centers on defining a vendor-neutral API abstracted from existing accelerator-attached file systems, with a modular backend that implements core functionalities including synchronous, batch, and asynchronous I/O operations strictly validated through comprehensive unit testing and synthetic benchmarks such as fil/filperf. To evaluate real-world efficacy, OpenDS is vertically integrated into leading AI inference frameworks like SGLang (HiCache) and vLLM (LMCache) and tested against critical workloads, including KV cache management, Retrieval-Augmented Generation (RAG), model weight loading, and Mixture-of-Experts (MoE) offloading. Ultimately, OpenDS establishes a robust, open foundation that promotes cross-platform innovation and accelerates the development of next-generation, storage-aware AI systems. Compared to traditional host-mediated I/O or vendor-locked direct memory access solutions, OpenDS significantly eliminates CPU bottlenecks and system memory overhead. By decoupling storage interfaces from specific hardware architectures, it delivers a universally applicable, low-latency data pipeline essential for scaling memory-bound AI deployments.
Biography
Nikhil B. Gaikwad is a Staff Engineer at Samsung Semiconductor Denmark Research (SSDR), where he focused on advancing AI infrastructure for high-performance, scalable, and energy-efficient inference systems. Prior to this, he completed three years of postdoctoral research at Aarhus University and Aalborg University in Denmark, working at the intersection of custom FPGA-based AI accelerators and GPU virtualization for large-scale AI workloads. He earned his Ph.D. from VNIT Nagpur, India, where his research centered on the hardware deployment of AI algorithms for edge intelligence.