Real-time prediction of heart attacks and arrhythmias

The aim of this project is  to detect and predict heart arrhythmias before they occur. Heart attacks and arrhythmias typically happen without noticeable warning. Predicting heart arrhythmias, such as atrial fibrillation, in real-time and with enough anticipation allows patients to seek immediate medical attention and prevent potential health complications. This project searches for subtle changes in heart dynamics, captured via ECG or PPG (via a smarthwatch, for example), using state-of-the-art machine learning tools. The project is funded by Luxembourg National Research Fund (FNR).

Critical transitions

Catastrophic events (i.e. sudden changes that affect systems stability) occur in various fields and at various levels. Examples include earthquakes, stock market crashes, and population extinction; for individuals, it is hypothesised that the onset of diseases such as cancer follows similar patterns. If we could understand the critical transitions (CTs) that induce catastrophes, we would be better equipped to prevent them arising or at least to mitigate their effects. The proposed research confronts this problem within a range of disciplines in the areas of clinical science, immunology, biology, and finance. It aims to classify CTs according to their general dynamical features and then to provide the foundations for:

a) identifying early warning signals to enable more timely, reliable predictions of catastrophes;

b) developing tools to model, analyse, and detect CTs in diverse areas of application.

Ultimately, the overall goal of the project is two-fold: to support more advanced research of CTs within scientific disciplines; and, in multiple fields, to improve society’s ability to anticipate CTs to undesirable states.The project entails eleven doctoral students and ten supervisors and it is funded by the FNR.

Next generation of deep brain stimulation (DBS)

Deep brain stimulation (DBS) is a surgical therapy for several movement disorders (e.g. Parkinson’s disease (PD) and Essential tremor) and psychiatric diseases. During this procedure, an electrode is implanted into the brain, constantly delivering electrical pulses to specific regions of the brain. This project aims to considerably improve the efficiency of DBS while reducing side effects. In particular, it develops algorithms to deliver personalised stimulations that adapt to the a patient’s needs, and computational tools to aid physicians initise DBS parameters. This interdisciplinary project involves collaborations with the Centre Hospitalier de Luxembourg (CHL) and University of Oxford. This project is funded by the FNR.

Closed loop DBS concept that uses multiple feedback sources to fine-tune stimulation

Network inference

The aim of this project is to understand how networks of genes and other molecules interact over time to produce disease- related phenotypic behaviours. Constructing reliable models of these regulatory networks allows us to simulate hypotheses that can be used to guide experimental testing, thus accelerating our understanding of how diseases function. Networks are inferred from time-series data. This also contributes to broader theoretical research into mathematical, algorithmic and software tools for understanding causal dynamic network structures.These tools have been applied to understand the mechanisms of action of circadian clocks and the immune system. The project aims to identify changes in biological network structures that occur when a system is perturbed in responses to stimulation, pharmacological intervention or genetic mutations. This project involves a collaboration with the Department of Plant Sciences at the University of Cambridge.

Predicting financial crashes

The project aims to identify hidden dynamics in financial systems and test potential early warning signals for critical transitions represented by mini-flash crashes in equity markets. Financial markets are known for their complexity due to the presence of many different types of agents whose actions belong to a very diverse set of strategies. Their collective behaviour in limit order markets results in time series of transaction prices for a particular security. The first part of the project studied causal effect of latency delays on occurrence of mini-flash crashes. This latency delay slows down the trading while reducing price impact, and results in decreased number of mini-flash crashes. The project aims at explicit descriptions of welfare gains due to usage of different order types, taking into account the microstructure of financial markets. The project is funded by the FNR.

Predicting the progression of Parkinson’s disease

The clinical presentation and rate of progression of PD vary considerably between patients. Our understanding of this heterogeneity is still very limited, preventing clinicians from making accurate predictions. This project pools longitudinal time-series data from seven PD cohorts in Europe and the US. The large quantity of data is unprecedented in this context. It allows us to build and evaluate advanced models of disease progression using state-of-the-art techniques from statistics and machine learning. We are particularly interested in clustering as a mean to redefine PD subtypes, which will then allow a more personalized diagnosis and prognosis. Future work will apply this methodology to other diseases, such as Alzheimer’s. This work is funded by the University of Luxembourg and it is a collaborative effort between the University of Cambridge, Newcastle University, the Academic Medical Center in Amsterdam and the LCSB.