Modalities
🧬 Targeted proteomics
🏥 Electronic healthcare records
⌚ Wearable data
🏠 Remote monitoring data
Machine Learning · Deep Learning · Clinical Data Science
I build deep learning and representation learning systems for clinical data.
PhD student at Imperial College London working across proteomics, electronic health records, and wearable data to build predictive, interpretable models for Parkinson's disease and related clinical monitoring problems.
Datasets
UK Biobank
Parkinson's Disease Progression Initiative
Global Neurodegeneration Proteomics Consortium
Minder Study
Publications
The Lancet eClinicalMedicine
Alzheimer's & Dementia
Tools
Live interactive demos
Hugging Face Spaces
Agitation Monitoring demo
Retention Rate demo
Profile
I work on modelling problems where the artifact matters as much as the idea: calibrated prediction, multimodal fusion, interpretable outputs, and tools that can be inspected through papers, demos, and case studies rather than abstract claims.
Projects
Selected projects and case studies.
Research spanning predictive modelling, deep learning, and clinical translation.
Plasma Proteomics and Machine Learning for Parkinson's Disease Prediction
Develop machine learning approaches for Parkinson's disease prediction from plasma proteomics and evaluate how the signal generalizes across datasets.
- Large-scale plasma proteomics data, with external validation in an independent cohort.
- Supervised machine learning, explainability analysis, calibration checks, and model error analysis.
- Interpretable risk modelling with attention to robustness and generalizability.
Multimodal EHR + Proteomics Fusion
Learn shared patient representations from structured EHR and proteomics to improve subgroup discovery and prediction.
- Longitudinal EHR plus targeted proteomics.
- Deep learning and representation learning for multimodal fusion.
- Ongoing work on stratification and interpretable latent structure.
Agitation Monitoring Demo
Detect and inspect agitation episodes in dementia using remote monitoring signals and an interactive demo.
- In-home monitoring and longitudinal behavioural sensing.
- Interpretable ML for episode detection.
- Publication in The Lancet eClinicalMedicine and a live Hugging Face demo.
Agitation Profiles in Dementia
Characterize heterogeneous agitation patterns from longitudinal in-home monitoring data.
- Repeated in-home monitoring measurements over time.
- Longitudinal profiling and interpretable analysis.
- Distinct agitation profiles reported in Alzheimer's & Dementia.
Flagship Project
Plasma Proteomics and Machine Learning for Parkinson's Disease Prediction
This project explores whether plasma proteomics can support Parkinson's disease prediction and whether machine learning models can recover clinically meaningful signal from large-scale biomedical data.
- Research question: can plasma proteomics contribute useful predictive signal for Parkinson's disease, and what can model behaviour reveal about heterogeneity in the data?
- Large-scale plasma proteomics data, with additional validation in an independent cohort.
- Supervised machine learning, explainability analysis, calibration checks, and error analysis.
- The project focuses on interpretability, generalizability, and understanding where models succeed or struggle.
- Overall, the work is aimed at informing more reliable biomarker-driven prediction tools.
What I Build
Core strengths across modelling, data fusion, and translation.
Work spanning multimodal integration, model development, evaluation across datasets and deployment.
Prediction and clinical stratification across heterogeneous health data
Combining targeted proteomics, electronic health records, and smartwatch-derived digital biomarkers to improve early prediction and subgroup definition.
Representation learning for structured and longitudinal clinical signals
Designing deep learning pipelines that learn useful latent structure from multimodal patient data while preserving interpretability and portability.
Cross-cohort evaluation and interpretable outputs
Evaluation spans calibration, explainability, and external validation across large-scale biomedical datasets.
Training
Technical formation across neuroscience, biology, and machine learning.
PhD, AI in Biomedicine and Healthcare
Imperial College London · 2024-2028
MSc, Translational Neuroscience
Imperial College London · 2022-2023
BSc, Biology
National and Kapodistrian University of Athens · 2017-2022
Oxford Machine Learning Summer School
University of Oxford · July 2024
Brain Omics 2.0
Human Technopole, Milan · November 2024
CCAIM AI and Machine Learning Summer School
University of Cambridge · September to October 2025
Tools
Selected demos and interactive tools.
Agitation Monitoring demo
A live tool for agitation monitoring, built from the same line of work as the dementia monitoring publications.
Retention Rate
A clinical trial retention toolkit and live demo focused on participant retention, retention-rate planning, and trial operations support.
Publications
Selected papers and presentations.
An interpretable machine learning tool for in-home monitoring of agitation episodes in people living with dementia
The Lancet eClinicalMedicine, volume 80.
- Proof-of-concept study on in-home monitoring of agitation episodes in dementia.
- Directly connected to the live Agitation Monitoring demo featured above.
Identifying diverse agitation profiles in dementia: Insights from longitudinal in-home monitoring data
Alzheimer's & Dementia, 20(S8).
- Longitudinal in-home monitoring data used to identify distinct agitation profiles.
- Companion output to the agitation monitoring tool and broader dementia work.
Contact
Get in touch.
marirena.bafaloukou22@imperial.ac.uk
Primary academic contact for collaboration and research enquiries.