Adapting the drug discovery pipeline for Fragile X Syndrome using Artificial Intelligence

Details

Supervisors

Verleysen, Michel ; Maignan, Nicolas

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil biomédical, à finalité spécialisée

Abstract

Fragile X Syndrome (FXS) is a neurodevelopmental disorder that affects cognitive and behavioral functions. In preclinical research, the Live Mouse Tracker (LMT) system provides valuable data on mouse behavior. This master's thesis aimed to evaluate the feasibility of developing machine learning (ML) models to differentiate FXS mouse models from healthy mice using the LMT data. Additionally, the study aimed to assess the explanatory power of the ML models and compare the usage of raw positional data versus behavioral event data detected by the LMT AI system. Two ML models were employed: a LightGBM model and a ResNet101 model. SHAP interpretability was applied to provide insights into the factors driving the models' predictions and to generate insights about the differences between the two groups. The results demonstrated that the ML models achieved performance metrics surpassing random chance. The models provided insights into the features that contributed to the predictions, such as mouse traveled distance, activity levels, and low-velocity regions. Furthermore, this thesis compared the usage of raw positional data and behavioral event data. Both approaches showed promise in differentiating the two groups, highlighting the potential of different data representations in analyzing LMT data. In conclusion, this thesis demonstrated the feasibility of ML models in differentiating FXS mouse models from healthy mice using LMT data with balanced accuracy above. The models exhibited predictive power and provided insights into the differences between the two groups with SHAP interpretability graphs. The comparison of data representations enhanced the understanding of LMT data analysis. These findings contribute to the field's understanding of FXS and offer valuable insights for using the LMT system in preclinical research for drug discovery in FXS.