IDENTIFICATION OF TRANSCRIPTOMIC CHANGES DURING BRAIN AGING AND NEURODEGENERATION USING INTERACTOME-BASED SUPPORT VECTOR MACHINE MODEL
01/29/2020
Tibor Nánási1, 2, István Ulbert1, 2, Tony Wyss-Coray3, 4, 5, 6, Benoit Lehallier3, 4, 5
1 Institute of Cognitive Neuroscience and Psychology, Research Centre for Natural Sciences, Budapest, Hungary
2 Faculty of Information Technology and Bionics, Péter Pázmány Catholic University, Budapest, Hungary
3 Department of Neurology and Neurological Sciences, Stanford University, Stanford, CA, USA
4 Wu Tsai Neurosciences Institute, Stanford University, Stanford, CA, USA
5 Paul F. Glenn Center for the Biology of Aging, Stanford University, Stanford, CA, USA
6 Department of Veterans Affairs, VA Palo Alto Health Care System, Palo Alto, CA, USA
Gene expression profiles vary greatly during aging and are associated with cognitive decline and neurodegenerative diseases. High-throughput measurement of the proteome is still technically challenging but transcriptomics can be used to infer proteomics changes. Here, we study human brain aging, Alzheimer’s disease (AD) and other neurodegenerative conditions using RNA-seq data available via the GTEx Consortium and Mayo Clinic. We hypothesized that analyzing gene expression changes in context of the protein-protein interactome (PPI) will lead to more conclusive biological findings than single gene analysis. First, a graph was constructed based on PPI catalogs optimized for high reliability (PICKLE, OmniPath). Second, condition-specific link weights were assigned as follows. For each PPI link, a Support Vector Machine (SVM) model was fitted to predict group affiliation (younger/control or older/disease) of the samples using gene expression data corresponding to the linked proteins. Predictive performance was quantified in a cross-validated model with Matthews Correlation to assign link weights. Finally, in the resulting weighted graph, proteins with unexpectedly high centrality were selected to test for Overrepresentation Analysis of Reactome terms. The approach was validated on repeated measurements within the GTEx data. Reproducibility of top gene sets increased and also, inclusion of PPI information to the model turned out to be essential to produce functional enrichments in a robust manner. Later stage of neocortical aging showed extensive similarities with AD compared to other neurodegenerative diseases. Overall, our results demonstrate the benefits of our novel integrative approach and emphasize key neuro-immunological and vascular aspects of brain aging and neurodegeneration.