Associate Director Astellas Pharma Weymouth, Massachusetts, United States
Disclosure(s):
Souvik Bhattacharya, PhD: No financial relationships to disclose
The tutorial session will be divided into two parts. The first part will cover fundamental machine learning concepts, including the challenges posed by high-dimensional and sparse data. It will provide an overview of common feature selection methods (filter, wrapper, and embedded methods) and discuss the importance of model interpretability using techniques such as SHAP (SHapley Additive exPlanations) values. We will emphasize the relevance of these concepts to typical pharmacometrics modeling scenarios, such as exposure-response analysis. The second part will be a hands-on session where participants will apply the techniques discussed in the first part to a realistic pharmacometrics dataset. Guided exercises will include data preprocessing, the use of various feature selection methods, and building predictive models to evaluate the selected covariates alongside model performance in R using a range of R packages (caret, shapviz, CORElearn, glmnet, randomForest, ranger, xgboost, mRMRe, Boruta, kernlab, pROC, corrplot). Participants will gain practical experience in interpreting model results and assessing the clinical or pharmacological relevance of the selected variables in specific contexts.
