Follow us
P076 – Stability and Influences on DNA Methylation Markers Used for Body Fluid Identification
The determination of the body fluid is important for forensic case contextualization. One common approach is the use of DNA methylation-based biomarkers that exhibit body fluid-characteristic patterns. DNA methylation is a central component of the epigenetic machinery and serves as a stable yet dynamic marker, displaying inter-individual variability.
In this study, potential influencing factors on DNA methylation of commonly used markers for body fluid identification were investigated and thereby the accompanying uncertainty in body fluid prediction. Because the effect of DNA methylation variance on prediction—and thus on the associated uncertainty—depends on the markers and the mathematical models employed, various marker combinations and modelling strategies were examined.
DNA methylation profiles of body fluid-characteristic markers were extracted from publicly available datasets, focusing on 450K and EPIC-array technologies. More than 700 datasets comprising raw data of over 90.000 samples were collected, content- and quality-filtered, normalized, and further processed to create combined datasets for downstream analysis. Whenever available, potential covariates such as technical variables, age, sex, disease status, lifestyle, and other environmental factors were considered as possible relevant traits and deeper characterized. The potential of Random Forest Classification and the foundation model TabPFN (Tabular Prior-data Fitted Network) for model development and investigation of the possible influence of traits on prediction accuracy was quantified. Particular attention was given to the underlying mathematical foundations of the models, with emphasis on uncertainty quantification, feature importance, and the potential changes introduced by the inclusion of covariates.
The results will be presented with emphasis on understanding and appreciation of the underlying variation (e.g. cell-type composition) and classification model used on body fluid identification.
Authors
- Jana Naue (Institute of Forensic Medicine, Medical Center - University of Freiburg, Faculty of Medicine, University of Freiburg, Germany)
- Laura Schmelzer (Institute of Forensic Medicine, Medical Center - University of Freiburg, Faculty of Medicine, University of Freiburg, Germany)
- Ferdinand Suchanek (Institute of Forensic Medicine, Medical Center - University of Freiburg, Faculty of Medicine, University of Freiburg, Germany)
- Peter Pfaffelhuber (Institute of Mathematics, University of Freiburg, Germany)
On the same topic
