Contact UsTechnology Highlights
Pioneering Graph-Based Relational Learning
Digitalization has vastly increased the amount of data and information available to us. While knowledge graphs let us connect data layers, our pioneering GraphAI technology combines logical reasoning and deep learning to discover and exploit relations between those data layers. This improves the performance of node classification, delivers link prediction, integrates multimodal and incomplete data sources and provides explainability for complex AI models.
What we learn from multimodal, unstructured and even incomplete data enables us to advance both fundamental and applied technology research. Our application areas include personalized immunotherapeutic cancer vaccines, development of new industrial materials, domain-specific large language models (LLMs) and generative AI.
Accelerating Vaccine Discovery with Software Optimization
Governments and scientists are in a race against time to identify and control the spread of emerging viruses before they become pandemics. Key to their success is the rapid development of new vaccines. NEC Laboratories Europe and NEC OncoImmunity (NOI) are working with the Coalition for Epidemic Preparedness Innovations (CEPI) to develop vaccines providing broad protection against SARS-CoV-2 variants and other betacoronaviruses.
The key to containing a potential pandemic is mass producing a safe and effective vaccine within the shortest amount of time. NEC Laboratories Europe is accelerating computational workloads and increasing the efficiency of NOI computational pipelines used for vaccine development. Using our AI acceleration technology, SOL, companies can dramatically reduce computation times, which can deliver results of analysis weeks or months earlier.
Accelerating High-Performance Computing and AI Systems with SOL
AI requires highly optimized and sophisticated AI frameworks. While many AI frameworks provide a simple programming model for scientists, they often fail to deliver peak execution performance. It is common for teams of performance engineers to re-implement algorithms from scratch when performance becomes critical: a long and costly process.
SOL overcomes this by accelerating the execution of neural networks, delivering optimal performance for parallel and vector-based computations and AI systems while, at the same time, preserving users original experience. Furthermore, SOL decouples neural network software from hardware, enabling optimal support for frameworks like TensorFlow and PyTorch on hardware platforms even when these are not supported in the original frameworks.
Advancing Simulations with Machine Learning
Reliable simulations require accurate descriptions of interatomic interactions. Machine-trained interatomic potentials have superior modeling properties compared to classical force fields, resulting in simulations of far greater accuracy. Despite this, they have yet to be widely adopted. One reason for this is system-specific datasets are required to train them, which are often not available. At NEC, we are leveraging active learning and uncertainty-driven molecular dynamics to bootstrap concise but comprehensive datasets, while limiting the use of computationally expensive quantum-mechanical solvers.
Reliable simulations require machine learning techniques to accelerate the estimates of molecular and material properties. One such improvement developed by NEC Laboratories Europe is a self-tuning Hamiltonian Monte Carlo approach – a hybrid simulation method consisting of elements of molecular dynamics and Monte Carlo. A simulation’s performance depends on the choice of its parameters. Our approach allows us to calibrate simulation parameters without external input, which eliminates the need for grid-searching. Through our advances, we have discovered atom-specific timesteps that further accelerate our simulations.
A New Frontier in Personalized Therapeutic Cancer Vaccines
Neoantigens can be used to develop personalized therapeutic cancer vaccines that teach the patient's immune system to recognize and kill cancer cells. The NEC Neoantigen Prediction Pipeline uses bioinformatics and machine learning to analyze the DNA and RNA of cancer patients to identify neoantigens originating from cancer mutations. NEC Laboratories Europe was a key contributor in developing the prediction pipeline, providing machine learning algorithms for the system, which is being used by NEC biotech partners for the research and development of immunotherapy.
The prerequisite for destroying cancer is the binding of T cell receptors to neoantigens on the surface of cancer cells. NEC Laboratories Europe, in collaboration with NEC Laboratories America, developed the AI model, Attentive Variational Information Bottleneck (AVIB). AVIB represents a first step towards predicting the likelihood that T cell receptors will recognize certain neoantigens. For cancer immunotherapy, this allows biotechnologists to select neoantigens that have a high chance of being recognized by the patient’s immune system, which increases vaccine efficiency.