B0981
Title: Prediction approaches for partly missing multi-omics covariate data: An overview and an empirical comparison study
Authors: Roman Hornung - University of Munich (Germany) [presenting]
Frederik Ludwigs - University of Munich (Germany)
Jonas Hagenberg - University of Munich (Germany)
Anne-Laure Boulesteix - LMU Munich (Germany)
Abstract: Multi-omics data, involving various types of omics data for the same patients, hold promise as covariates in automatic outcome prediction, with each omics type potentially contributing unique information to enhance predictions. However, one of the challenges faced is block-wise missingness - where different omics data types are not available for all patients in the training data and the data on which the automatic prediction rules are to be applied, the test data. This is referred to as block-wise missing multi-omics data. An overview of existing prediction methods designed to tackle such data is provided. Subsequently, the results of a benchmark study based on a collection of 13 publicly available multi-omis datasets comparing the predictive performance of several of these approaches for different block-wise missingness patterns are presented. A discussion concludes on the results of this empirical comparison study, leading to some tentative conclusions.
