(23g) Multi-Omics Integrative Analysis Using Mutual Information-Based Machine Learning

Systems-scale analysis of multiple layers of molecular and cellular data has significant potential for both industrial and biomedical advances. While approaches have been developed and applied extensively for mining of a single type of genome-scale data at a time, the direct integration of multiple levels of these data types has remained challenging. Post hoc integration of individual analyses can provide some supporting insight, but direct integration of data types, whether driven by a priori knowledge or the data itself, could have significant impact. We present here a unique approach to the analysis of a longitudinal multi-omics dataset; while the specific application is in the context of a non-human primate model of malaria, the underlying approach could be used for arbitrary applications with available multi-comics datasets. We analyzed relationships across multiple biological layers using a mutual information-based machine learning approach to integrate heterogeneous longitudinal datasets and constructed an atlas of multi-omics relatedness networks (MORNs). Using this technique, we were able to detect signatures with substantial biomedical relevance. To our knowledge, this is the first report of large-scale, longitudinal multi-omics analysis of malaria in any system, as well as a novel approach for multi-omics data integration.