A new Nature study by researchers from VUB and KU Leuven challenges the reported decline in scientific disruption. Using VSC’s high-memory CPU nodes to analyse citation networks with hundreds of millions of connections, the team found that missing reference data and a visualization bug had exaggerated the trend.

Can AI reliably detect freezing of gait across different Parkinson’s disease cohorts? Researchers tested a deep-learning model using wearable movement sensors across seven cohorts. The results show why expert oversight remains essential and how limited cohort-specific data can improve model performance. VSC computing resources enabled the extensive multicentre evaluation and fine-tuning experiments.

Predicting the stability of crystalline materials is essential for developing advanced materials and pharmaceuticals. Researchers developed a new thermodynamic integration framework that calculates Gibbs free energies entirely under constant-pressure conditions, improving accuracy while simplifying the workflow. The method was validated using ice polymorphs and CsPbI₃ and relied on more than 10,000 GPU hours on VSC infrastructure.

Researchers at VUB developed a patient-relevant orthotopic lung cancer model and combined SEPARATE-Seq with spatial transcriptomics to map the immune landscape of lung tumours. Supported by VSC's high-performance computing infrastructure, the study provides new insights into immune cell behavior and offers a valuable resource for cancer immunology and future therapy development.

Researchers at Ghent University used large-scale simulations on the Flemish Supercomputer Center (VSC) to investigate how dangling bonds influence the electronic structure of III-V quantum dots. Their findings reveal how surface defects affect semiconductor performance and identify promising materials for next-generation infrared sensors, imagers, and optoelectronic devices. This research advances the design of safer, high-performance quantum dot technologies.

Researchers at Ghent University developed a patient-centered approach to detect cancer signals hidden in blood plasma using cell-free RNA (cfRNA). By combining RNA sequencing with large-scale computational analysis, they identified personalized molecular patterns that distinguish cancer patients from healthy individuals, advancing the future of precision oncology and personalized diagnostics.

Adaptive design of experiments improves model identification by selecting the most informative tests while reducing overall testing time. Demonstrated on gearbox power-loss modeling, the approach balances parameter accuracy with efficient test sequencing. Supported by VSC computing resources, it enables faster, reliable model development with fewer experiments.

Researchers at KU Leuven leveraged VSC Tier-1 supercomputing to train a large EEG foundation model on over 8 million segments, enabling robust brain signal decoding across tasks and subjects. The work underpins an ICLR 2026 study and achieved first place in the NeurIPS 2025 EEG Challenge, demonstrating the impact of large-scale AI and high-performance computing in neuroscience.

This study benchmarks single-cell chromatin accessibility technologies for training deep learning models in regulatory genomics. Using large-scale datasets from mouse and Drosophila, it shows that increasing dataset size can offset lower per-cell coverage. The optimized HyDrop v2 protocol enables cost-efficient, high-quality data generation, supporting robust sequence-to-function modeling powered by high-performance computing.

Brain activity measured with EEG offers a non-invasive way to study cognitive function. This research uses machine learning and brain connectivity patterns to detect Alzheimer’s disease and frontotemporal dementia, two conditions that are difficult to distinguish clinically. The approach improves diagnostic insights while highlighting the role of brain networks in neurodegenerative disorders.