Detection of Erythropoietin from Blood Biomarkers
Specialeforsvar: Jiayu Li
Titel: Detection of Erythropoietin from Blood Biomarkers
Abstract: Blood doping is widely recognized as a form of cheating that undermines the principles of fair competition and equality in sports while also posing potential health risks to athletes.
Common blood doping techniques include blood transfusions and the use of erythropoiesispromoting hormone drugs, such as recombinant erythropoietin (rhEPO). Given the challenges and expenses associated with direct testing methods, our approach emphasizes the utilization of indirect and closely related direct blood biomarkers to determine whether athletes have engaged in blood doping with rhEPO.
In the majority of detection methods, population-based thresholds are employed as criteria for identifying potential cases of blood doping. While this approach has proven highly effective in detecting cases of blood doping among athletes, it is not without its limitations. Occasionally, clean blood samples from individuals with naturally higher baseline values of blood biomarkers may be incorrectly flagged as potential blood doping cases. This misclassification occurs due to the reliance on population-based thresholds rather than personal ones, neglecting the unique characteristics and variations among individuals. To address this issue, it is crucial to not only establish population-based thresholds but also explore personal thresholds.
This thesis employs both traditional statistical techniques, such as multiple linear regression, and machine learning classification methods, including multinomial logistic regression and random forest. These methodologies are utilized to develop a predictive classification model using experimental data collected from participants in both sea-level and high-altitude environments. The primary goal of these models is to detect the usage of rhEPO in blood samples through indirect blood biomarkers.
In the pursuit of establishing individualized thresholds, the baseline values of blood biomarkers are incorporated into the model. Furthermore, to improve the effectiveness of classification for each model, we integrate essential physical indicators from the experimental participants, along with key direct biomarkers. This provides a comprehensive assessment of blood doping,
addressing both individual-level and population-level considerations. The findings suggest that, regardless of the various methods employed, detecting rhEPO in blood samples has yielded unsatisfied results. The classification accuracy consistently falls within the range of 60% to 70%, with only minor variations observed in the criteria used to distinguish between the rhEPO and placebo groups. However, the use of personalized thresholds has demonstrated an improvement in the accuracy of rhEPO detection, which is crucial for minimizing errors. Additionally, the integration of machine learning methods and an expanded of blood biomarkers has shown promise in achieving higher classification accuracy. In future research, the collection of more relevant data and the inclusion of additional blood biomarkers may further enhance the effectiveness of classification.
Vejleder: Helle Sørensen
Censor: Christian Damsgaard Jørgensen, Aalborg Universitet