Machine learning empowered prediction of consequences following intraventricular hemorrhage in preterm neonates using targeted proteomics

Aktivität: Vortrag ohne Tagungsband / VorlesungPräsentation auf einer wissenschaftlichen Konferenz / Workshop


Preterm neonates with intraventricular hemorrhage (IVH) are at risk to acquire posthemorrhagic ventricular dilatation (PHVD). We performed targeted proteomics analysis on different biological matrices, urine and blood, derived from a longitudinal retrospective neonatal patient cohort including six timepoints collected over the last 9 years. We employed explainable machine learning (ML) algorithms to identify those patients with IVH who are prone to develop PHVD, predict survival and to identify disease-specific protein-biomarkers. This unique setting enabled us to detect potential predictive biomarkers that could help in therapeutic decision-making and parental counselling.
In recent years targeted proteomics has developed into a powerful protein quantification tool in biomedical research, systems biology and clinical applications. The targeted approach applied in this study was a Proximity Extension Assay (PEA), which combines the specificity of dual antibody recognition with qPCR readout and therefore enables the detection of low concentrations of proteins in biofluids as previously shown in several publications. We applied dissimilar ML methods from different domains (statistics, regularization ML, deep learning, decision trees, Bayesian) to the targeted proteomic data for biomarker detection to gain insight into the development of PHVD. We evaluated 600 models trained with different algorithms for the prediction of patients who are prone to develop PHVD and trained 600 models to predict the survival of IVH patients. Each model was submitted to a 10-fold cross-validation repeated 10 times each. We were able to identify previously known, like NEFL (neurofilament light chain) and novel potential biomarkers, as key indicators for the development of PHVD and the prediction of survival.
The purpose of our study was to provide new insights, detect clinically relevant biomarkers and explore the differences between the patient groups within clinically defined time frames. The main strengths are the rigorous evaluation of the calculated ML models and the access to an exceptional neonatal patient cohort.
Zeitraum20 Sept. 2023
Ereignistitel15th ÖGMBT Annual Meeting: Life sciences and cutting-edge technologies
OrtSalzburg, ÖsterreichAuf Karte anzeigen

Research Field

  • Molecular Diagnostics


  • Bioinformatics
  • Bioinformatik
  • Machine Learning
  • Artificial Intelligence
  • Molecular marker
  • pediatrics
  • health
  • proteomics