One manner in which cheating athletes attempt to gain an advantage in endurance sports is through blood doping, most commonly with the use of erythropoietin (EPO).

EPO is a hormone, naturally produced by kidneys, that stimulates the production of new red blood cells. Exogenous EPO preparations were developed for the treatment of various forms of anemia but can also be abused by doping endurance athletes to increase oxygen delivery to muscles. While EPO can be detected with targeted analytical methods, deciding which athletes to test and the timing of tests remain critical questions for the successful application of this method.

Within the project “EPO-Predict,” DFKI research group Smart Service Engineering lead by Prof. Dr.-Ing. Wolfgang Maaß will support WADA by applying various machine learning or rather deep learning techniques to predict whether an athlete has taken an EPO drug or not. Besides the identification of the use of EPO, the project further targets the following outcomes:

(1) cost reduction by optimizing over a subset of input features,
(2) better use of available blood data for target analysis of EPO,
(3) improvement of specificity of tests, and
(4) Bayesian optimization over feature subsets.

First, several blood indicators and a set of questionnaire items regarding the athlete’s recent activity have been used as a dataset for training. To achieve the mentioned objectives, DFKI plans to use several machine learning models, including various deep learning models. Within the first workshop at WADA’s European regional office in Lausanne, DFKI started to analyze available data.

contact for scientific information:
Prof. Dr.-Ing. Wolfgang Maaß
Head of Research Group Smart Service Engineering
German Research Center for Artificial Intelligence (DFKI)
E-Mail: Wolfgang.Maass@dfki.de

idw 2019/04