Novel computational strategies to deconvolute co-occurring conditions in Down syndrome

PROJECT SUMMARY The triplication of chromosome 21, which encodes ~225 genes, is the cause of Down syndrome. Individuals with Down syndrome experience a wide and variable spectrum of co-occurring conditions. Though we know the cause, which is trisomy 21, we have a poor understanding of how and why this chromosomal abnormality drives associated co-occurring conditions. A better mechanistic understanding of these connections will provide the basis for not only improving the care of individuals with Down syndrome but also for the general population. Research efforts such as the Human Trisome and INCLUDE Project are generating multi-omics data on large cohorts of individuals with trisomy 21 to gain such mechanistic insights. Given the complexity of the problem – the dosage increased of ~225 genes connected to a wide spectrum of conditions – existing tools for -omic data analysis struggle to leverage this information properly and separate generic from context-specific cellular responses. We must be able to analyze these data in the context of the full disease spectrum that individuals with DS experience, from genes to proteins to pathways. Further complicating these analyses, we and others have shown that certain genes and pathways are hypersensitive to perturbation, thus we often identify generic responses through standard analysis methods, when our goal is to find disease- and context-specific changes. These hyperresponsive genes and pathways obscure context-specific signals. We propose to develop the methodology to find the context-specific signal associated with trisomy 21. Our first aim is to develop methods to identify shared genetic mechanisms between complex diseases and molecular changes in Down syndrome co-occurring conditions. We will leverage genomic and transcriptomic datasets from a wide array of previously collected association studies. Our second aim is to develop a method to separate generic from context-specific signals in -omic datasets. We will employ a novel generative neural network simulation to generate different disease contexts, for which individual -omic samples can be compared. Our third aim is to determine the molecular connections between chromosome 21 genes and the co-occurring condition using search over knowledge graphs. All methods developed will be made public through the INCLUDE Data Coordination Center platforms.