Current Grant
08/24 – 02/26
Dysferlinopathy is a genetic muscle condition with an autosomal recessive inheritance produced by mutations in the DYSF gene that encodes for the dysferlin protein [1]. As well as is seen in other muscular dystrophies, disease progression in dysferlinopathy is heterogenous. In the first paper published using COS-1 data [2], authors arrived to the conclusion that the rate of disease progression is variable and not linked to the initial phenotype [3]. They described that within the group of patients that had been symptomatic for at least 30 years, 1/3 were mildly or moderately affected, and a similar proportion of mild to moderate affected patients was seen in the group that had been symptomatic for less than 17 years, suggesting the presence of disease modifying factors [2]. Since that first publication, some other factors have been identified to influence disease progression, such as intense physical activity during teenage years [4] and higher baseline water T2 values in skeletal muscle MR scans [5].Our main aim is to identify factors that could predict disease progression and prognosis in dysferlinopathy patients. We are interested in exploring if the combination of genetic, clinical, biochemical and imaging variables could predict disease progression in dysferlinopathy by using artificial intelligence. This approach will help us to create algorithms to predict disease progression and identify what is the average weight of each individual variable in the final algorithm.
To achieve this, we will work with the clinical, genetic, biochemical and imaging data obtained from the patients enrolled in COS1 natural history study. We will assess changes in these variables over time and classify patients into fast or slow progressors compared to the mean progression rate of the whole cohort. Then, we will create a machine-learning based pipeline to deal with data missingness, data re-sampling and/or data augmentation and automatic feature selection. The resulting data will be fed into a machine learning algorithm, proven to outperform other methods on medical datasets [12]. We will optimize, train and evaluate the model in a nested cross-validation to ensure that no bias is introduced during any of the steps and ensuring the reliability of the performance estimation. We will apply explainable AI techniques to ensure the predictions of the Machine Learning model are transparent and to extract knowledge from the final model that can be by researchers and clinicians.
This project is being led by Prof. Jordi Diaz-Manera, working along Jose Verdú Diaz and Dr Carla F Bolaño Diaz at the John Walton Muscular Dystrophy Research Centre.
[1] Aoki M, Takahashi T. Dysferlinopathy. In: Adam M P, et al ed. GeneReviews((R)). Seattle (WA), 1993.
[2] Harris E, Bladen CL, Mayhew A, et al. The Clinical Outcome Study for dysferlinopathy: An international multicenter study. Neurol Genet 2016;2:e89.
[3] Moore U, Gordish H, Diaz-Manera J, et al. Miyoshi myopathy and limb girdle muscular dystrophy R2 are the same disease. Neuromuscul Disord 2021;31:265-280.
[4] Moore UR, Jacobs M, Fernandez-Torron R, et al. Teenage exercise is associated with earlier symptom onset in dysferlinopathy: a retrospective cohort study. J Neurol Neurosurg Psychiatry 2018;89:1224-1226.
[5] Moore U, Caldas de Almeida Araujo E, Reyngoudt H, et al. Water T2 could predict functional decline in patients with dysferlinopathy. J Cachexia Sarcopenia Muscle 2022;13:2888-2897.
