Post-doctoral Research Scientist School of Biomolecular and Biomedical Science, University College Dublin, Ireland
Abstract: Recent advancements in artificial intelligence (AI) including machine-learning (ML) and other modelling methods, shows that these will have had a profound impact in medicine, especially in treating diseases with various clinical phenotypes and non-specific symptoms. For example, utilising heterogeneous patient data, AI methodologies have the potential to predict the onset of disease or disease progression, improved the accuracy of disease diagnosis, and inform choice of therapy more accurately.
From a wealth of translational OMICs data available both in-house and from publicly-accessible databases, we have focused on modelling diseases where diagnosis is challenging and cumbersome and symptoms poorly predict outcomes. The first of these diseases is pre-eclampsia, a serious complication affecting one in every 12 pregnancies. Annually, it claims the lives of 50,000 mothers and 500,000 babies, making it the world’s deadliest pregnancy complication. At present, birth of the baby is the only treatment and safest for the mother. Pre-term delivery is associated with a significant risk of long-term neurodevelopmental infant morbidity and mortality, and accounts for a huge proportion of admissions to neonatal intensive care unit. Clinical diagnosis remains extremely challenging and ‘rule-in’ diagnostic tests are an urgent need. We have used powerful machine learning algorithms to combine patented blood-based biomarker signals with selected maternal haematological/demographic/clinical assessment data, to aid clinical evaluation in real-time in 250 sick pregnant women. Such a decision support tool, once validated, will help to prevent unnecessary adverse outcomes for mother and baby.
We also modelled platelet lipidomics and miRNomics data from 76 patients with cardiovascular disease (CVD) to identify important features involved in the progression of stable cardiovascular disease to acute myocardial infarction (MI). CVD is a progressive disease and remains the leading cause of morbidity and mortality worldwide. Although clinical risk factors for adverse cardiovascular outcomes are well-defined, biomarkers to prognosticate disease progression and identify high-risk individuals are still lacking. Using our algorithm, we have identified several blood enzymes and lipids that collectively discriminate the patient with stable CVD versus MI. As a proof on concept, we identified troponin, a well-established biomarker for MI, as the most important variable to correctly classify patients. Evaluation of our set of features over time may identify vulnerable patients likely to experience adverse outcomes and open novel avenues to explore the underlying pathophysiological derangements in the progression of cardiovascular disease.
Collectively, the use of AI on biomedical OMICs data will not only enable the development of novel risk stratification tools and disease biomarkers but also help to advance the understanding of molecular mechanisms underlying diseases, ultimately shaping the future of healthcare across disciplines.