My PhD
Title:
Applying novel machine learning techniques to understand the relevance of biomarkers in rare diseases.
Stakeholder:
Amicus Therapeutics
The Research:
To date, there are over 7000 different rare diseases that afflict millions of individuals in the World and are responsible for the deterioration of physical health, mental health, and socioeconomic conditions. These conditions remain a mystery, with around 26% of rare disease patients not surviving beyond the age of 5, suggesting a gap in the literature with no formal diagnostic criteria nor a known cure. Thus, there lies an urgent need to build knowledge on the clinical manifestation of these conditions and to reduce the patient's time to diagnosis.
The initial focus of the PhD is on a disease known as Fabry (E75.21), a rare lysosomal storage disorder affecting multiple body systems. The disorder may lead to heart disease, kidney disease, strokes, and painful peripheral neuropathy, in addition to several other problems. Although it is X-linked, it can cause serious implications in women as well as men. The symptoms of a classical Fabry patient include painful hands and feet (acroparesthesia), a characteristic rash (angiokeratomas), and abdominal symptoms; in the nonclassical form of the disease, these features may be mild or absent, and the condition usually becomes clinically apparent later in life than in the classical form. As the manifestations are relatively non-specific and/or look clinically similar to other diseases, it can be difficult to diagnose and patients may visit an average of 7 doctors and wait for over 15 years before being diagnosed with Fabry disease. This sequence of specialist visits generates a series of misdiagnosis, which leads to increased patient anxiety, delay in receiving treatment or even incorrect treatment being administered, and worsening of the patient’s disease-related manifestations. Those with an incorrect diagnosis can also miss out on trials of a new treatment, such as that of Substrate reduction therapy. Moreover, the lack of a large patient pool hinders the ability to obtain a common clinical manifestation of the condition.
The project branches out to tackle three different areas of research:
- Applying Reinforcement Learning to understand the relevance of cardiac biomarkers in the early diagnosis of Fabry disease - This stem aims to uncover a series of makers that may indicate an early diagnoses of Fabry in patients with cardiovascular manifestations.
- Detection of Fabry patients through application of machine learning techniques on primary care EHR data - Due to the lack of secondary care data in many rare disease patients and with the aim of identifying contributing factor to diagnosis at the earliest possible stage. This stem aims to prompt suspicion of a possible Fabry diagnosis mainly using primary care data with addition of some secondary care data, if necessary.
- Applying Machine Learning techniques to monitor Fabry disease progression in renal patients This stem focuses on assessing renal Fabry patients improvement or deterioration, through the analysis of biomarker data, which may lead to a better understanding of both pre- and post- diagnostic biomarkers.
The delay’s in the procurement of Fabry patient data have driven the projects focus onto developing a novel AI modelling technique that thrives in imbalanced scenarios. Currently, Deep Learning techniques combined with Reinforcement Learning are paving the way to improved detection of rare events.