Doctoral Candidate Position 15 - National and Kapodistrial University of Athens, Greece
AI pipeline development for metabolomics data analysis - Application to Parkinson's studies
Alberto Moreno obtained his bachelor's degree in Physics from the University of
Murcia in 2023, where he specialized in climate data analysis and completed a
thesis on fire danger synchronicity in Europe, work that was later presented at
the EGU General Assembly 2023 and resulted in a co-authored paper in
Environmental Research Letters (2025) on extreme fire weather synchronicity in
Europe.
He then pursued a Master's degree in Artificial Intelligence at the Universidad
Internacional de Valencia (2023–2024), carrying out his thesis at the University
of Cologne on transfer learning and explainable AI to predict Alzheimer's
disease conversion.
From October 2024 until he joined the BICEPTS Marie Sklodowska-Curie
Doctoral Network, he worked at the University of Murcia as a research assistant
on the project Evolution of Fires at Global and Regional Scales in Response to
Climate Forcing, contributing to manuscripts on climate–fire interactions. In
parallel, through his collaboration with the DICSO research group on the
Erasmus+ project Accept the Challenge: Gamification in Online Higher
Education, he also delivered outreach talks to education students on the role of
science in society.
ORCID: 0009-0006-1386-5195
Description of project
Detecting Parkinson's disease (PD) is a devastating neurodegenerative disorder characterized by diverse symptoms where accurate diagnosis remains an unmet challenge. Biomarkers-based PD diagnosis can potentially overcome the limitations of traditional clinical observation-based approaches if it become accurate and explainable. This research aims to develop explainable Artificial Intelligence methods based on machine and deep learning, integrating flexible feature selection, model selection, sample classification, and results interpretation based on metabolomics features for PD diagnosis and classification purposes.
Approach:
Step 1: Utilize cutting-edge machine and deep learning methods to perform feature selection and ranking, model selection, implement explainable classification, and employ advanced visualization methods. Develop a versatile and transparent end-to-end pipeline for sample classification, particularly well-suited for scenarios with a high feature-to-sample ratio and noisy features.
Step 2: Establish robust and efficient computational strategies for predictive modeling and implement them using state-of-the-art Python technologies. Consider combining metabolomics data with other modalities to improve predictive accuracy.
Step 3: Validate by creating predictive models for distinguishing early and late-stage PD using pertinent patient samples.
Expected Results:
(i) Develop a comprehensive workflow that implements the end-to-end explainable AI pipeline for metabolomics data analysis, granting researchers and clinicians transparent disease insights. (ii) Illuminate disease mechanisms and the evolving role of metabolite sets in Parkinson's progression, potentially offering novel strategies for delaying or halting disease advancement.