Professional photo

Kyriakos Soulios

PhD Candidate - Data Scientist

Helmholtz Centre for Environmental Research - UFZ
Department Computational Biology
Permoserstr. 15, 04318 Leipzig, Germany


Building: Building 4.1
Room: Room 237
Email: kyriakos.soulios@ufz.de


Research interests

  • Computational toxicology
  • Graph Neural networks
  • Unsupervised / Semi-Supervised learning
  • Contrastive learning
  • Molecular Representations
  • Uncertainty Quantification
  • Conformal Prediction



Scientific Network

ORCID

Google Scholar


Current Projects

Toxicity prediction mining the chemical universe.

The goal of the project is to provide accurate and trustworthy ML models which can predict the toxicity of chemical found in the environment. The larger scope of the project consists of:

  • Prioritize the toxicity assessment of chemicals in the environment
  • Monitor/Regulate the suspected substances
  • Design "green" chemicals

The second leg of the project focuses more in the trustworthiness of the developed ML models by integrating Uncertainty Quantification methods and exAI techniques and analysing into the actual predictions and explanations.

Keywords: Computational toxicology, Graph Neural Networks, Molecular Property prediction, Uncertainty Quantification


Academic Career

09/2021 - Current: Doctoral Researcher in Bioinformatics

Department of Computational Biology, Helmholtz Centre of Environmental Research

09/2021 - Current: PhD Candidate in Bioinformatics, Faculty of Mathematics and Computer Science, University of Leipzig, Germany

10/2015 - 02/2021: Integrated Master's in Cheminformatics, Faculty of Pharmacy,  National and Kapodistrian University of Athens, Greece


Publications

  1. Soulios, K., Scheibe, P., Bernt, M., Hackermüller, J., & Schor, J. (2023). deepFPlearn +: enhancing toxicity prediction across the chemical universe using graph neural networks. Bioinformatics, 39.

Grants & Awards

  1. Helmholtz Institute for Data Science Trainee Network Grant: 6000€ for a 3-month research stay working on Topology-guided GNNs for toxicity prediction