3S Project-II: Assessment of Subsampling-based Convolutional Neural Network (SCNN) algorithm for possible use in anti-doping procedures related to the athlete biological passport.

Code: 24T01WM

As sample swapping is an approach to evade doping detection, the use of machine learning-based algorithms has the potential to improve accuracy of fraud detection by improving the sensitivity of detection of urine exchange through improved targeting of DNA identification (i.e. “DNA fingerprinting”), which can be used to demonstrate the exchange of an athlete’s urine with that of another individual. Therefore, this project aims to integrate the already developed algorithm for flagging sample swapping into anti-doping protocols [3], using it both "retrospectively" and "prospectively", and to analyze the passport assessment process to evaluate the impact of the resulting tool in the anti-doping community.

This will be accomplished by providing a standalone and user-friendly tool for testing by APMUs, which will generate a score indicating the degree of similarity between one sample and all other samples within an anonymized longitudinal profile of an athlete. In order to develop the tool in a targeted and beneficial way, an understanding of its impact on the passport assessment process as well as of potential changes through the tool is required. Therefore, the project will entail several stages, following a design science research methodology, covering the following phases:

(1) exploring and understanding the passport assessment process, e.g., through online interviews with Athlete Passport Management Units (APMUs), (2) design and development, i.e., updating the algorithm to provide a score of similarity between a given sample and the other samples in an athlete's longitudinal profile, (3) demonstration i.e., the algorithm will be accessible through a standalone and user-friendly software, enabling domain experts like APMUs to easily access and test the algorithm, (4) evaluation of the impact on passport assessment by the APMUs, using anonymized longitudinal profiles and (5) communication, i.e., presentation and publication. By implementing this project, the capability of anti-doping organizations to identify instances of sample swapping by athletes shall be significantly enhanced. Through the integration of a machine learning-based algorithm, domain experts will be able to make more informed assessments of steroid data, leading to more accurate and effective antidoping efforts. The project will evaluate how, and under which conditions experts use the tool and how it influences their assessment. For this, a combination of quasi-experiments and qualitative or quantitative questionnaires will be applied. Through observations and semi-structured interviews within the quasi-experiment, we aim to analyze tool usability, confidence, and accuracy in steroid data assessment as well as potential changes in productivity through the machine learning-based tool. This allows drawing conclusions on the impact of tool support in sample swapping on the anti-doping community. A validated tool ready to use by the anti-doping community will be delivered.