CoMPaSS-NMD - Computational Models for New Patients Stratification Strategies of HNMD

Hereditary Neuromuscular Disorders (HNMDs) are inherited genetic diseases that cause progressive muscle weakness and atrophy, significantly impacting mobility and often leading to long-term disability [1, 2]. These conditions dramatically impair social participation, result in loss of independence, and can lead to premature death [3]. Beyond affecting daily activities and social engagement, HNMDs necessitate extensive care from others, imposing substantial personal, social, and economic burdens on patients, their families, and society. HNMDs are a subgroup within the broader category of musculoskeletal disorders, which are the leading cause of years lived with disability. Although individually rare, HNMDs collectively affect over two million people worldwide [4].

Diagnostics for HNMDs often involve multiple costly investigations that can be partial or inconclusive, leaving individual prognoses uncertain. Current genetic studies have identified 16 groups of HNMDs with over 600 causative genes [5]. Classification of HNMDs is now primarily based on factors such as mode of inheritance, genetic mutation, clinical presentation of weakness, histological features, age of onset, rate of progression, and prognosis.

Molecular DNA analysis has become the standard for diagnosing these diseases. However, extensive genetic sequencing has revealed a broad phenotypic spectrum, with different clinical presentations in patients or healthy relatives carrying mutations in the same gene, complicating the correlation with clinical "Gestalt." Patients with HNMDs often exhibit similar features and changes in muscle imaging and histology, yet some may have mutations in different genes, while others may have no obvious detrimental variants. These observations highlight the limitations of current diagnostic approaches and suggest that inherited neuromuscular disorders may result from complex, largely unknown biological interactions.

Consequently, 60% of patients presenting with neuromuscular disease symptoms do not receive an accurate molecular diagnosis, and effective treatments are currently unavailable for most of these diseases [6].

In recent years, Artificial Intelligence (AI), particularly Machine Learning (ML), has matured to a level that supports its application in the life sciences, paving the way for personalized healthcare [7]. ML-based computational tools offer an unprecedented opportunity to expand medical knowledge by providing new insights into the pathogenetic and transmission mechanisms of HNMDs, achieving precise clinical characterization and diagnosis, supporting accurate clinical decisions, and improving future prognosis, which is often challenging to predict in clinical practice.

The CoMPaSS-NMD project aims to develop, apply, and test novel ML tools for the stratification of HNMD patients, with the following four general objectives:

1. Generate robust and reliable datasets to develop computational tools based on validated data.
2. Create clinical data integration solutions for the classification of clinical phenotypes to support patients, healthcare professionals, the scientific community, and individuals potentially affected by HNMDs.
3. Develop evidence-based guidelines to improve patient management based on phenotype stratification, enhancing the current standard of care for healthcare providers.
4. Create an online platform for public data collection, compliant with the European Union (EU) FAIR (Findable, Accessible, Interoperable, Reusable) principles, for patients affected by neuromuscular diseases, known as ATLAS-NMD.

The CoMPaSS study processes both pre-existing and newly obtained data in two parts: (i) a retrospective observational study using genetic, histopathology, and Magnetic Resonance Imaging (MRI) data collected from centers in Great Britain (UNEW), France (CERBM), Finland (SFF), and Italy (FSM); and (ii) a prospective study involving the collection of clinical, genetic, histopathology, and MRI data from 500 previously undiagnosed patients at clinical reference centers in Italy (FSM, UNIMORE) and Germany (LMUM).

Each participant in the prospective study will undergo:

• A standardized clinical evaluation protocol assessing all muscle districts, with data recorded in an electronic clinical record (eCRF).
• Muscle MRI performed based on clinical indications.
• Muscle biopsy of an affected muscle, if clinically indicated.
• Genomic analysis of DNA extracted from peripheral blood lymphocytes. Participation in the study will be granted only after the participant or their legal representative has reviewed the study information sheet, provided informed consent for the processing of sensitive data and privacy, and signed the consent document.

The study employs computational tools for clustering based on unsupervised multidimensional processing, which uses the internal structure of data to define similar groups among patients. Regarding data privacy, data minimization, and data access rights, the study follows a Federated Learning approach. This ensures that models for stratification and classification can be trained collaboratively by several entities while keeping data decentralized at each clinical center.

The study adheres to a patient-centered co-design methodology, involving strong stakeholder participation and networking with other project consortia, while respecting all principles of data protection and management.

Regarding the expected outcomes, the study aims to provide researchers with effective solutions for integrating health data and more precise classification of clinical phenotypes. Based on these findings, evidence-based guidelines will be developed using the stratification of distinctive clinical traits, ultimately offering a superior standard of care for the diagnosis and prognosis of patients with HNMDs.

The study is funded by the European framework program "Horizon Europe" under the call HORIZON-HLTH-2022-TOOL-12-two-stage.