Analytical Data Review on an Artificial Intelligence Platform for Doping Control in Horse Racing.

This study explores the use of artificial intelligence (AI) to improve the efficiency and accuracy of reviewing chromatograms for doping control in horse racing, where over 90% accuracy was achieved in the initial evaluation of the AI model.

Artificial Intelligence in Doping Control

• The research revolves around the use of AI for doping control in horse racing. This process involves screening prohibited substances (PS) in horse biological samples using gas chromatography/mass spectrometry (GC/MS) and liquid chromatography/mass spectrometry (LC/MS). The screening generates a sizable number of chromatograms, which pose a challenge in manual evaluation.
• Recent developments in AI give the opportunity to utilize image recognition tools to classify these chromatograms as either “positive” or “negative” for PS presence, thus potentially increasing the efficiency and accuracy of these evaluations.
• The study carried out uses an image recognition tool in “Alteryx Designer,” a data analytic software, to train a predictive model with over 6000 manually classified chromatograms. The aim is to use this trained model to perform initial analysis of chromatograms.

Efficiency and Accuracy of AI in Doping Control

• To gauge the model’s reliability, around 700 manually classified chromatograms were analyzed by the model. The model showcased over 90% prediction accuracy, marking it as a potentially useful tool for initial chromatogram analysis.
• The model’s application on two in-house screening methodologies covering over 300 drug targets proved its robustness. Here, it classified “suspicious,” “positive,” and “negative” chromatograms with high accuracy, with no occurrence of false negative classification.

Challenges in Implementation of AI in Doping Control

• The study acknowledges two primary challenges in integrating the developed model into regular testing. Firstly, the AI model’s runtime on a standard office PC falls between 3 to 5 hours. It implies that the analysis time is a factor that could hamper the practical implementation of the AI model.
• Secondly, the developed model’s compatibility with various tools poses a concern. The model’s success in its current form relies on data extraction from specific instruments and software, making it inflexible for use across a broader spectrum where a variety of instruments and processing software may be involved.