Identifying New Disease Genes & Mechanisms for Musculoskeletal Disorders in 100K Genomes Project using Bioinformatics, Phenotyping & Machine Learning

Genetic disorders which affect the development of the skeleton or muscles are collectively common, even if individually rare. Providing a genetic diagnosis for the patients and their families is important for ending what is often a lengthy diagnostic odyssey. For their clinicians, it may inform provision of the correct treatment. Understanding the genetic basis of these rare musculoskeletal (MSK) disorders may also provide insights into common MSK disorders, which are a major cause of disability and impairment of quality of life for millions of people in the UK.
In the past, genetic diagnosis of rare MSK diseases has relied on sequencing panels of known genes to identify the causative gene, but the diagnostic yield of such panel-based sequencing is low because many disease genes have not yet been identified.
With technological improvements and cost reductions, sequencing of patients' entire genomes (the full complement of their DNA) has become a possibility. Furthermore, many types of genetic variants can be interrogated from genome sequence data, not just those involving single base pairs, but also more complex duplications, deletions or transpositions of segments of the genome as well as variants in the regions between genes - the introns. These regions have increasingly been recognised to play important roles in regulating gene expression but we have considerably less understanding about their clinical significance.
Interrogation of patients' genomes to identify the disease-causing variants therefore still presents many challenges. Recognising the potential of this genome sequencing approach, the UK launched a national programme (100KGP) to identify pathogenic variants in 100,000 patients, with the aim of improving diagnoses for these patients that might also inform their personalised treatment. Run by Genomics England, sequencing of these patients is now complete and it is estimated that diagnoses have been found for a quarter of the rare disease patients so far. Solving the rest of these cases will require intense effort on behalf of the research community to investigate the different variant types described above.
This proposal aims to contribute to that effort focusing on patients with musculoskeletal and related developmental conditions.
We will use both existing GeL algorithms and our own bioinformatics tools to analyse the genome sequence data to ensure we have investigated all possible variants, and then employ a variety of genetic strategies to assess whether the genes are potentially pathogenic. We invariably need additional clinical or x-ray data to that already collected by the GeL programme. However, this is often available in medical records so we have identified routes to retrieving this which involve clinicians and patients themselves. We have established a clinical multi-disciplinary team to enable discussion of cases, and will employ expertise in clinical radiology assessments to ensure systematic analysis of x-ray data. We will also ask patients to provide us with self-reported data, as we know from other research studies that patients are very good at remembering which bones they have broken and when. Finally, we will see if machine learning or 'artificial intelligence' can help us identify patterns in these vast and complex datasets which could not be identified by our manual inspection.
We anticipate that these efforts will help us provide diagnoses for many more patients in the 100KGP and can then be adopted for other diseases in the 100KGP providing genetic diagnoses for many more patients.

Whole genome sequencing (WGS) has the potential to revolutionise diagnosis of Rare Diseases. Recognising this, the UK has established a national programme to sequence 100,000 genomes (100KGP). To date, analysis in 100KGP has primarily focused on known disease genes for a given condition and on particular variant types - predominantly single nucleotide variants (SNVs) and a diagnostic yield of ~25% has been achieved. A more research-focused effort is now required to investigate novel disease genes and variant types, such as copy number, other structural variants (CNVs/SVs) and non-coding variants that are largely unexplored in the 100KGP to date. In order to address this requirement, we will focus on patients with musculoskeletal (MSK) disorders in the 100KGP which are a clinically and genetically heterogeneous group of conditions accounting for >1,000 cases in 100KGP. In preliminary studies, we have already identified a non-coding variant that contributes 1% to diagnostic yield of osteogenesis imperfecta and 2 complex SVs in known genes.
We will use novel bioinformatics techniques to comprehensively analyse the WGS data, integrating the various variant types to identify putative novel disease genes. We will combine this with deep phenotyping as core MSK clinical data has not been collected by 100KGP and is required in the assessment of candidate genes. We will also evaluate whether machine learning can be used to identify clusters of genotypes or phenotypes from these complex high dimensionality datasets enabling novel genotype/phenotype correlations to be observed.
Although this proposal focuses on MSK conditions, we anticipate that evaluation of the bioinformatics algorithms, platforms for deep phenotyping at scale and machine learning approaches will be informative for other disease domains in the 100KGP and can be leveraged to increase diagnostic yield across the dataset, as well as helping to maximise the research potential of this unique resource.