COMPASS-NMD: Structuring uncertainty in neuromuscular disease

Under the coordination of Rossella Tupler at the University of Modena, COMPASS-NMD is confronting that challenge by combining deep clinical phenotyping with genomics, imaging and digital infrastructure to deliver more precise classification, earlier diagnosis and meaningful prediction – laying the foundations for a more structured, data-driven space in neuromuscular care.

Across Europe, thousands of people live with neuromuscular diseases (NMDs) that do not necessarily shorten life dramatically, but profoundly reshape it. The burden is physical, social and economic, and for many patients the most difficult part is not only the management of the symptoms they suffer, but the uncertainty they live with.

It is against this backdrop that the COMPASS-NMD project was conceived. Funded under Horizon Europe, the project brings together clinicians, geneticists and computational scientists across Europe to rethink how neuromuscular diseases are diagnosed, classified and ultimately understood. Rather than focusing on genetics alone, COMPASS-NMD is building a new framework that integrates deep clinical phenotyping with genomic, imaging and histological data to transform uncertainty into structured, actionable knowledge.

“We started this project after many years working with patients with neuromuscular diseases,” explains Rossella Tupler of the University of Modena and coordinator of COMPASS-NMD. “These diseases severely impair the quality of life and although most patients do not die from them, they live very difficult life: these diseases progress over time, and patients may lose their independence, eventually becoming unable to walk or  care for themselves. In general, they encounter many difficulties in movement and they struggle to participate fully in work and social life.”:

Yet for many, these burdens are compounded by something less visible but equally disruptive: the long and often inconclusive path to diagnosis. Despite decades of genetic discovery, clarity in diagnosis remains elusive. Since the first Duchenne muscular dystrophy gene was identified, DNA analysis has become central to clinical practice. “Diagnosis today is largely based on genetic testing,” Tupler explains. “However, despite this, the majority of people remain undiagnosed because no clearly significant gene is identified.

“At the same time, neuromuscular diseases are strikingly heterogeneous,” she continues. “Some people have a very slowly progressing disease; others have a very severe form that begins early and leads to wheelchair dependence by the age of 20. This is completely different from someone who develops symptoms in their 50s.”

This variability does not simply describe different clinical trajectories; it shapes entire life paths. Life expectancy and quality of life can differ dramatically depending on when symptoms begin and how quickly they progress. For some, independence is lost early; for others, decline is slower but no less disruptive. In all cases, the consequences are felt well beyond the diagnosis, carrying social and economic implications that are impossible to ignore. “We must be frank: a disabled person who cannot move easily may have difficulty working, and there is a cost associated with this. There is also a burden on caregivers, so the social costs are very high.”

Against this backdrop of uncertainty about how the disease progresses in different individuals, the COMPASS-NMD team came to a clear conclusion: the core issue is not simply genetic complexity, but clinical classification. As Tupler explains: “From my experience, having seen thousands of patients, I realised that it is very important to classify what we call the phenotype, that is, the way the disease presents.”

This, she argues, is “crucial first for classification and second for prediction.” Because without understanding how a disease presents in detail, it is impossible to anticipate how it will unfold. The questions patients ask are immediate and deeply personal: “If a person is 30 and begins to have difficulty lifting their arms, what will happen next? Will they be able to run, walk, climb stairs? Parents ask me these questions when they want to have children: what will happen? These are very big questions.”

Deep phenotyping

The project was, therefore, built around a deliberate shift in emphasis. Rather than relying primarily on genetic sequencing and attempting to interpret it in isolation, the core idea was to standardise deep phenotyping with whole-genome analysis, MRI and histology.

“Our approach is based on standardised deep phenotyping, combined with whole-genome analysis, MRI, and histology,” Tupler explains further. “MRI is a key instrumental tool for defining the clinical aspects of these myopathies, while muscle biopsy histology remains essential for subdividing patients into clinically meaningful groups.”

This integrated approach allows the consortium to do more than assign labels. “By taking all these aspects together, we believe that if we can stratify patients properly, we can better understand who may respond to therapy, how the disease develops and potentially how to stop it.”

The project also builds on data generated in earlier Horizon-funded initiatives, reworking them within a new framework of structured clinical phenotyping. Working with computational scientists, the team has developed artificial intelligence-driven algorithms to analyse multimodal datasets – clinical assessments, genomic sequencing, MRI imaging and muscle biopsy histology – identifying patterns across them.  The model analyses all four domains simultaneously, searching for clusters of shared characteristics that define clinically meaningful subgroups. Now, a new cohort of 500 individuals with myopathy is being studied across these same four domains. Data from their deep phenotypic and harmonised description, integrated with genomic, imaging ang histological pictures, will serve to validate the model, testing whether the algorithm can reliably stratify patients and generate clinically useful classifications.

For Tupler, this shift in approach was cultural as much as technical. “What is important is making a paradigm shift. Clinicians must start considering the phenotype in depth. You cannot rely only on DNA. In the same family, people can carry the same mutation but show no disease. This is a major challenge.”

Structured intelligence
One of the fundamental problems COMPASS-NMD addresses is fragmentation – not a lack of data, but a lack of integration. Over the past decade, previous projects have amassed large collections of MRI scans, genomic sequences or muscle biopsy data. Yet these datasets were often generated in isolation and, crucially, without standardised, structured clinical descriptions. As a result, researchers could analyse images or genes but struggled to link them reliably to how a patient’s disease actually presented and progressed. “If you do not have deep phenotyping, how can you correlate what you see in MRI or DNA with the patient’s actual condition?” Tupler asks. “Deep phenotyping is missing in most studies and that is the critical point we took on.”

To resolve it, the team developed a comprehensive clinical evaluation tool, the Structured Clinical Report Form, containing over 200 structured items. Clinicians select predefined options, with interpretation being standardised. As Tupler puts it, “There is no ‘fantasy’ interpretation. It is essentially binary: it is present or it is not.” By transforming subjective clinical observation into harmonised, structured data, deep phenotyping becomes the anchor that allows imaging, genomics and histology to be meaningfully aligned. This structured input enables computational scientists to build robust algorithms on coherent datasets rather than disconnected fragments. The result is not only diagnostic refinement, but predictive modelling.

ATLAS: a European repository for neuromuscular intelligence
At the heart of this digital infrastructure lies ATLAS, a centralised repository designed not simply to store data, but to integrate and structure it in a way that makes meaningful clinical interpretation possible. “We developed ATLAS as a repository where all anonymised and pseudo-anonymised data are collected in a standardised way,” Tupler explains. “Each patient is assigned a code linking clinical examination, MRI, DNA and histology. Data are then analysed at the Silesian University of Technology, in Gliwice, Poland, with results returned to ATLAS.”

In effect, ATLAS becomes the operational core of the project’s approach, the point at which deep phenotyping, imaging, genomics and histology converge. By ensuring that every dataset follows the same structured logic, it transforms what were once isolated streams of information into a coherent system capable of classification and, ultimately, prediction.

The long-term ambition for this approach extends beyond the life of the project. “ATLAS is designed to become a repository for the wider community. In the future, once validated, other clinicians could add patients and this would create a centralised European resource.”

Tupler draws a comparison with the US cancer repository, The Cancer Genome Atlas or TCGA, which has grown over time into a cornerstone of oncology research. In Europe, she observes, sustained infrastructure funding remains more difficult. “There is no stable mechanism to support them long term.” Yet without continuity, infrastructures like ATLAS, designed to turn structured data into structured intelligence, cannot fully realise their potential. And without that stability, data-driven transformation in health systems remains fragile.

Shortening the diagnostic odyssey
For patients, the implications of this work are immediate. Neuromuscular diseases are often characterised by prolonged diagnostic journeys, with many years of referrals and uncertainty. The project hopes that this can now change. A clinician could input a new patient’s data into ATLAS and receive an informed classification. “The aim is that, by inputting a patient’s data into ATLAS, a clinician can receive an accurate interpretation quickly, and this should improve diagnosis.”

Over time, the predictive power of these interpretations will only strengthen. If patients with identical genetic profiles are identified at different ages, likely progression trajectories can be inferred. While not equivalent to a full longitudinal study, such modelling offers meaningful clinical insight.

The broader impact on therapy development is equally significant. “Stratification is essential for therapy development,” says Tupler. “If you do not stratify patients, treatment effects may be diluted.” By identifying homogeneous subgroups, the platform supports biomarker discovery, response prediction and more targeted clinical trials. Even though COMPASS-NMD itself is a foundational research initiative, it provides a structured entry point for companies developing therapies.

As Tupler points out, the value lies in order: “It creates an ordered system, like organising books in a library so you can find what you need.”

Beyond the science
For Tupler, the implications of COMPASS-NMD extend beyond improved diagnosis and classification. They also touch on how neuromuscular diseases are understood within health systems and public policy. Reflecting on earlier registry work, she explains: “For years I ran the national registry for facioscapulohumeral muscular dystrophy. We calculated how much the state spends on it, while we also understood how much money could be saved, how budgets could be reduced, with better physical therapy.”

She also highlights the links that exist between social conditions and clinical outcomes. “We realised that the people with lower education, lower income, are more affected,” she says. “They often receive later diagnosis and reduced access to care, and this contributes to worse progression.”

Improved classification and earlier diagnosis have the potential not only to reduce uncertainty for patients, but to influence how care is delivered and resources are allocated, easing pressure on families, caregivers and health systems alike.

“These are things that the politicians and the policymakers should know about,” adds Tupler. “Structured data and better stratification are not abstract scientific exercises; they are tools that should inform decision-making and I would like to have a channel with the policymakers.”

Training and the next phase
Alongside these policy implications, COMPASS-NMD also reflects a deliberate model of interdisciplinary collaboration. Geneticists, clinicians, computational scientists, legal and ethics experts work side by side. Precision medicine, Tupler argues, demands new forms of dialogue. “Clinicians must explain challenges to engineers; computational scientists must explain their algorithms.”

For her, this exchange is essential. Genetics has evolved dramatically over the course of her career, and clinicians must now understand genomic individuality as routine. At the same time, computational scientists must grasp clinical complexity and uncertainty. The consortium even convenes roundtables with philosophers, a recognition that prediction in medicine carries ethical as well as technical dimensions. As Tupler puts it, the aim is to contribute to “the generation of the new generation of scientists.”

The project has now entered its validation phase. The cohort of 500 patients with neuromuscular disease who currently have no diagnosis is being recruited. For each, clinical, genomic, MRI and histological data are collected, harmonised and analysed within ATLAS. By the project’s conclusion, the expectation is that clinicians will be able to input structured data and receive meaningful classification – the first operational proof of principle for the COMPASS-NMD model.

Ultimately, however, the project’s ambition is human as much as scientific. “The ultimate goal is to improve quality of life, to reduce confusion and fear, to shorten the diagnostic odyssey, and to enable better follow-up and participation in society,” concludes Tupler.

If successful, COMPASS-NMD will have demonstrated that deep phenotyping, digital integration and structured analysis can turn fragmented datasets into clinically useful knowledge. But the remaining question is sustainability. As Tupler notes in relation to European infrastructures, “There is no stable mechanism to support this work long term and for initiatives like ATLAS, only continuity will determine whether proof of principle becomes lasting transformation.”