The talk starts at 12:15.

Please note that due to COVID-19, the participants can watch the streamed talk on Teams with a link (below).

Speaker: Pål Vegard Johnsen

Location: Only via Teams; click here to join on Teams.

Title: Explaining risk-of-disease models with genetic data using SHAP and subSAGE: How to evaluate the importance of the features?

Abstract:
SHAP values have become a popular procedure to interpret complex machine learning models, such as evaluating which features are important, as well as interaction effects. However, there are several pitfalls, such as understanding how SHAP values should be interpreted, not accounting for correlation between the features, as well as not accounting for the uncertainty in the SHAP values. In addition, the computational resources acquired is challenging for high-dimensional data.

 In this talk I will present a concrete example of applying SHAP values for a risk-of-disease model using high-dimensional SNP array data in order to measure feature importances as well as interactions. Results and reflections will be presented. I will also present an alternative method for not only measuring feature importance, but also inferring its uncertainty, using subSAGE.

Welcome!
Best regards,
Thea Roksvåg and Lars Henry Berge Olsen.