A0564
Title: Enhancing time-to-event prediction with high-dimensional omics data using exclusive Lasso regularization
Authors: Dayasri Ravi - Technical University Dortmund (Germany) [presenting]
Andreas Groll - Technical University Dortmund (Germany)
Abstract: The integration of high-dimensional omics data into survival prediction models has gained significant attention due to the growing availability of these datasets. Traditionally, a Cox regression model is employed, concatenating various omics data types linearly. Given that much of the omics data may be redundant or irrelevant, feature selection through penalization is often necessary. A notable characteristic of these datasets is their organization into blocks of distinct data types, such as methylation and clinical data, which requires selecting a subset of variables from each group due to high intra-group correlations. The proposal is to utilize exclusive Lasso regularization in place of the standard Lasso penalty. Exclusive Lasso promotes intra-group sparsity via the $L_{1}$-norm while encouraging inter-group selection through the $L_{2}$-norm, ensuring the selection of at least one variable per group. A significant challenge arises from the non-differentiability of the $L_{1}$-norm within groups, which is addressed by transforming this norm using quadratic and sigmoid function approximations to achieve differentiability. This transformation facilitates the use of a straightforward Newton-based approach to solve the intricate optimization problem. The methodology is applied to real-life cancer datasets, demonstrating enhanced survival prediction performance compared to the conventional Lasso-penalized Cox regression model.
