Research

Who we are

We are a research group at the University of Fribourg investigating how cognitive training reshapes brain function and behaviour. Our work spans fundamental neuroscience and applied intervention research, with a common thread: understanding the mechanisms of training-induced plasticity and translating that understanding into tools that improve health outcomes.

From neurological rehabilitation to reward and craving

The lab's origins are in neurological rehabilitation and the study of executive function plasticity. Our early work investigated how repeated practice on executive control tasks induces neuroplastic changes in healthy and clinical populations, using neuroimaging, neurostimulation, and pharmacological approaches to characterise the metaplasticity of inhibitory control, cognitive flexibility, and working memory. This foundational programme established that executive training produces measurable changes in underlying neural networks, not merely improvements on the trained task, and that individual differences in baseline neural organisation predict who benefits.

Over the past decade, this expertise has converged on a specific and consequential question: can executive training, particularly motor inhibitory control training, be used to modulate the brain reward system and reshape cue-driven craving? This question sits at the intersection of cognitive neuroscience, learning theory, and public health, and it now defines the core of our research programme.

What we study

Our current work is organised around three interconnected lines of research.

We study how motivated learning endophenotypes in Pavlovian conditioning and other reinforcement-learning strategies (i.e., sign-tracking vs. goal-tracking and model-free vs. model-based learning) shape vulnerability to cue-driven overconsumption and responsiveness to cognitive training interventions. This line provides the individual-difference framework that guides the personalisation of our interventions.

We develop and test response training interventions, using our gamified app-based platform to conduct preregistered, double-blind trials of Go/NoGo and approach-avoidance training across food, alcohol, cannabis, and tobacco domains. Our emphasis is on real-world effectiveness, with clinical partnerships that embed our interventions in existing care pathways and public health programmes.

We build digital behavioural biomarkers, most notably the Motivational Salience Index (MSI), which uses machine learning to infer cue-induced craving from implicit behavioural signals captured during smartphone-based cognitive tasks. The MSI provides the objective monitoring layer that connects our fundamental research on craving mechanisms to adaptive, closed-loop intervention delivery.

These three lines are mutually reinforcing. The endophenotype research tells us who will respond to training. The biomarker research tells us when and how much someone is craving. The intervention research tests whether training works, in whom, and under what conditions. Together, they form an integrated programme aimed at producing application-oriented research with tangible impact on disease prevention and treatment.

Our studies continue to rely on neuroimaging (high-density topographic EEG with electrical neuroimaging), neurostimulation, and pharmacological approaches, as well as their combinations, alongside the behavioural and computational methods that support our digital health research.

Technology transfer: BeweLab SA

Since 2020, our translational efforts have been structured through BeweLab SA (bewe.com), a University of Fribourg spin-off based at Biopôle Lausanne. BeweLab develops and commercialises the digital tools that emerge from our research, including the gamified training platform used in our intervention trials and the MSI behavioural biomarker technology (patented).

The relationship between the lab and BeweLab operates under formal conflict-of-interest management. All studies are sponsored and conducted by the university. Preregistration, data collection, analysis, and reporting are performed by the academic team, with analysis plans locked before unblinding. BeweLab does not access identifiable participant data; only aggregated results and scientific specifications are shared after peer-reviewed publication. This model ensures that the research remains independent while providing a structured pathway for validated scientific findings to reach the patients, clinicians, and public health systems that need them.

BeweLab currently collaborates with insurance partners (CSS, Groupe Mutuel, VIVA/Visana), clinical centres (CHUV, CPNO, Ensemble Hospitalier de la Côte), medtech and food industry partners, and a network of independent nutrition professionals. The company has been selected as laureate of the Biopôle Clinical Top-Up investment programme to conduct clinical trials evaluating the training platform as a companion to GLP-1 receptor agonist therapies.

Methods

Our research combines large-scale applied trials, computational approaches, and neuroimaging within an integrated methodological framework.

App-based randomised controlled trials at scale

A central methodological commitment of the lab is to test our interventions under conditions that are both scientifically rigorous and ecologically valid. We conduct preregistered, double-blind randomised controlled trials in which participants complete gamified cognitive training on their own smartphones over multi-week periods, with real-world behavioural outcomes tracked through ecological momentary assessment.

Our custom-developed training platform, built over five years and validated across multiple trials, supports adaptive difficulty algorithms, gamification mechanics, real-time compliance monitoring, and parametric experimental manipulations, all delivered at scale without requiring laboratory visits during the training phase. This infrastructure allows us to run adequately powered studies (typically N = 150-1,000) while maintaining the experimental control needed for causal inference. The platform has been adopted by independent research groups internationally (Prof. Craske, UCLA; Prof. Coppin, University of Geneva), confirming its robustness and versatility beyond our own research context.

All confirmatory trials adopt the Registered Report format, with Stage-1 peer review of hypotheses, design, and analysis plans before data collection begins.

Machine learning and computational modelling

Our work on digital behavioural biomarkers and intervention optimisation relies on supervised machine-learning methods applied to high-frequency behavioural data. We extract fine-grained features from smartphone-based cognitive task performance (reaction time distributions, error dynamics, sequential effects, response variability) and use these to train predictive models of craving and motivational states.

Our modelling pipeline encompasses regularised linear models, tree-based ensembles (XGBoost, LightGBM), and deep learning architectures (recurrent neural networks, multimodal fusion), with systematic benchmarking through cross-validation, ablation studies, and feature-importance analyses. A core concern is deployment readiness: we prioritise model architectures that are lightweight enough to run on mobile devices, ensuring that validated models can be integrated into real-world applications without prohibitive computational costs.

This computational work is conducted in collaboration with the Swiss Data Science Center (SDSC) and benefits from expertise within the lab in both experimental design and quantitative modelling, ensuring that the features we engineer are grounded in cognitive neuroscience theory rather than purely data-driven.

Electrical neuroimaging

To characterise the neural mechanisms underlying training-induced plasticity and affective learning, we use high-density topographic EEG (128 channels) analysed with electrical neuroimaging methods. This approach, developed over decades by the Lausanne school of EEG analysis (Michel & Murray, 2012) and applied extensively in our lab (Hartmann et al., 2016; De Pretto et al., 2019; Sallard et al., 2018), goes beyond conventional single-electrode event-related potential analysis by treating the full scalp topography as the unit of observation.

Our core analytic tools include Topographic Analysis of Variance (TANOVA), which tests whether experimental conditions engage qualitatively different neural generator configurations; Topographic Analysis of Covariance (TANCOVA), which identifies time windows where scalp topography maximally covaries with external variables such as craving indices or learning phenotypes; Global Field Power (GFP) analysis, which quantifies response strength independently of spatial configuration; and distributed source estimation (LAURA/sLORETA), which models the intracranial generators underlying scalp-recorded effects.

This methodology allows us to dissociate two fundamentally different types of neural change: modulations in the strength of a given network's activation (quantitative plasticity) versus reorganisation of the networks themselves (qualitative plasticity). This distinction is critical for understanding whether an intervention merely enhances existing processing or genuinely rewires how the brain evaluates reward cues, a question at the heart of our research programme.

We complement EEG with neurostimulation (transcranial electrical stimulation) and, in selected protocols, pharmacological manipulations, enabling causal inference about the role of specific neural systems in training-induced behavioural change.

Open science commitment

We are strongly committed to open science practice, and this commitment is embedded in the design of every study we conduct. Since adopting the Registered Report (RR) publication format as our default mode of confirmatory research, we have published over 11 Registered Reports, making our lab one of the most consistent adopters of this format internationally.

The Registered Report route ensures methodological rigour and optimisation through peer review before data collection begins; best research practice by controlling for publication bias and questionable research practices; reproducible methods and analyses through detailed, pre-committed protocols; and replicable results through adequately powered studies with pre-specified analytic strategies. All data, analysis code, and study materials are made publicly available on the Open Science Framework (OSF).

We view open science not as an administrative requirement but as a scientific commitment. In a field where effect sizes vary widely and failed replications are common, locking analytic decisions before seeing the data is the most credible way to build cumulative knowledge. It also means that our findings inform the field regardless of whether the results are positive or negative, which is essential for honest progress in intervention science.

Funding

Our work is supported by public and private grants, notably from the:

• Swiss National Science Foundation (SNSF)
• Swiss Foundation for Alcohol Research
• Velux Foundation
• Nestar Foundation
• Novartis Foundation for Biological and Medical Research
• Swiss Parkinson
• UniFR tech transfer