The analysis of big datasets with Artificial Intelligence (AI) methods in the broadest sense is fundamental to further progress in life science and medicine, from personalized therapies over drug discovery and medical image analysis to providing infrastructures and predictive models for health and pandemics data. Our research provides tailor-made AI-based algorithms and solutions for this.
ScaDS.AI Dresden/Leipzig develops and applies AI technologies to predict the structure of undruggable targets, building structural models of disease variants in precision medicine, design antibodies and vaccines including COVID-19. Our center will study small molecules as well as biologics, such as antibodies and vaccines. Furthermore, we will develop AI technologies for protein target identification, protein structure prediction, variant characterization for precision medicine, ultra-large library screening to identify new drug molecules, as well as engineering of broadly neutralizing antibodies against viral infections and vaccine candidates designed to elicit these. The success of AlphaFold2.0 combined with the rapidly increasing structural and sequence databases spur research in this field. Specifically, we will test different types of convolutional neural networks for structure prediction and drug design.
We were among the first to use AI technologies for decoding functional Magnetic Resonance Imaging (fMRI). Our center has strong expertise in data-driven analysis of large-scale, high-resolution 4D imaging experiments and e.g. developed a content-adaptive image representation. Pioneering the interface between AI and neuroscience, we will use AI as a conceptual framework to study the brain (AI to neuroscience) and to improve AI by insights from how the brain works (neuroscience to AI). We will e.g. develop unsupervised learning algorithms for Spiking neural networks (SNNs) incorporating synaptic plasticity and dynamics (MSSM) to detect consciousness in Completely-locked-In-Status or Coma patients. We will continue to push the limits of biomedical imaging using AI, e.g. in reconstructing ultra-high-resolution, multiparametric MRI, and building deep, generative models of patients.
Infrastructure and devices
We are part of the NFDI4Health consortium building standards and infrastructures for managing and sharing data, derivatives, and models based on clinical and epidemiological studies, such as Covid-19. We develop infrastructure design for medical Internet of Things (IoT) systems in operating rooms, smart and context-aware biomedical technologies, AI-based clinical assist systems, and AI-based clinical decision support systems. We will design infrastructure for medical IoT systems in rural areas, reference IoT implementation in laboratories, or 5G/6G applications in healthcare. We integrate our applications into clinical practice considering regulatory aspects according to MDR and IVDR. We provide new concepts to integrate and represent data in clinical decision-making processes.
Models from limited data
Our center develops new AI such as sparse inference frameworks, methods to achieve perfect adaptation for learning solutions of partial differential equations, or generalize Newton and Lagrange interpolation to arbitrary-dimensional spaces. Applications include epidemiology or high-dimensional omics data. Models will be developed for cancer therapy, infectious diseases, metabolic diseases, and cardiovascular disease. We will adapt AI to multi-parametric molecular data in precision medicine e.g. on- and off-target prediction and biomarkers for personalized imunomodulatory therapies or model-based ML approaches that overcome limitations in genome and transcriptome assembly. We e.g. develop the first transcriptome-wide expression profile assessing prostate cancer risk by Next-Generation-Sequencing, models of retinal nerve fiber layers or whole 3D body scans, and a formal ontological framework.
Our strength is the development and application of AI technologies within realistic clinical settings. We will translate precision medicine concepts into clinical practice, explore pathogenic mechanisms beyond genomics, and pursue their integration into clinical decision making. We will develop AI-based, distributed methods for diagnosis support that continuously improve as data in hospitals (MII) and studies (NFDI4Health) become available. We collaborate with experts across ScaDS.AI Dresden/Leipzig.
To train the next generation of leaders at the interface of life science and medicine with AI, we recruit international experts to ScaDS.AI Dresden/Leipzig, such as Humboldt Professors Sayan Mukherjee and Jens Meiler. Furthermore, we work closely with the German Medical Informatics Initiative (MII).