| Summary: | By using a comprehensive online resource generated by the PsychENCODE Consortium for the adult brain, Wang et al, have identified and embedded
functional elements, quantitative-trait loci (QTLs), and regulatory-network linkages into a comprehensive deep-learning model, in order to predict
psychiatric phenotypes from genotypic and transcriptomic data. The end-result is a biologically-relevant Deep Boltzmann Machine architecture connecting
genotype, functional genomics, and phenotype data, with conditional and lateral connections that improve trait prediction over traditional additive
models. In the present thesis we are focusing on schizophrenia and follow a different route, by using the regulatory relationships including the enhancers, transcription factors (TFs), miRNAs, and target genes (TGs) in this resource as priors to different techniques: For example, we are not using the TG-TG linkages as lateral connections and TF-TG linkages as conditional connections to the same network, but we include them as prior information to network analysis and to denoising autoencoders (DAE), respectively. Our main goal is to implement different methods by exploiting different priors each time and finding the overlapping genes supported by as many methods as possible.
|
|---|
