Anti-doping analysis is a crucial measure to fight against cheating and doping activities in sports. The Athlete Biological Passport is widely implemented, including its two modules, namely Haematological and Steroidal. WADA's anti-doping controls involve the collection of blood and urine samples. In the case of urine samples, WADA's executive committee has found in their recent investigation that the exchange of urine samples is possible whereby a doping athlete can exchange their urine with another individual's urine to evade a positive test. This means an efficient way for dopers to pass through an anti-doping test is to swap their urine sample, providing a clean sample from another person instead. During an analysis of urine samples, the steroid profile is quantified for each sample, consisting of 11 markers of endogenous steroids.
In the suspicious case of sample swapping, the anti-doping organisation with testing authority confirms it by performing DNA analysis across multiple samples, which is targeted to specific samples based on risk factors such as the physical demands of a specific sport, the number of positives previously in that sport/country, and/or athlete specific intelligence. Candidate selection for additional testing should be such that it maximises the chances of discovering dishonest athletes while conforming to an anti-doping organisation's budgetary limit. Therefore, there is a need for an adaptive model which can flag the sample substitution by the athlete. In other words, a flag that is based on the likelihood that a given sample belongs to the same individual. As the steroid profile consists of 11 markers, it is hypothesised that an Artificial Intelligence (AI)-based tool embedded in pattern classification applied on this high-dimensional profile has the potential to improve the detection of sample swapping.
In this project, we expand the research ambit of the collaboration to improve the ability to uncover sample swapping by developing a pattern recognition/classification algorithm that provides a score of similarity of one sample steroid profile with all others provided by the same athlete. Some of the major goals include studying the steroid profiles of athletes, finding the best indicators through statistical-based methods, deploying state-of-the-art deep learning algorithms, and developing a fully-fledged pipeline model for the detection of swapped samples.
Partner
Deutsches Forschungszentrum für Künstliche Intelligenz (DFKI) GmbH, World Anti-Doping Agency (WADA), Deutsche Sporthochschule Köln (DSHS)