Neural networks for extraction of multi-omics modules
||Neural networks for extraction of multi-omics modules|
||SNIC Small Compute|
||Hendrik Arnold De Weerd <firstname.lastname@example.org>|
||Högskolan i Skövde|
||2022-09-01 – 2023-09-01|
The human interactome is a complex cobweb that spans multiple biological domains. Correspondingly, we now have an arsenal of high-throughput technologies that can measure the domains of the central dogma of molecular biology – and beyond. Indeed, DNA, DNA methylation, chromatin availability, mRNA expression, and protein abundance can, and have been, extensively measured, and these data are readily available to the public. In the case of complex diseases, i.e. maladies that arise from multiple factors, these measurements have elucidated several disease-related elements in multiple cellular domains. In other words, complex diseases have been found to have several disease-associated genetic variants, differentially expressed genes, and differentially expressed proteins.
Despite the massive amounts of data, the effect of a disease-associated change in one domain – such as a genetic mutation - is not easily predicted on a different domain, such as gene expression. There are at least two reasons for this difficulty. First, tools to analyse the impact of disease associated changes across the interactome have lagged behind. The reasons for this are plenty, with the non-linear relationship between cause and effect of disease changes being a prominent obstacle. Second, paired data that measure several omic’ domains from a sample are relatively sparse, especially from a perspective of modern data-intensive analysis tools.
This project aims to utilize deep learning techniques to elucidate causal variants in a shared latent space where two cellular domains are encoded in a double autoencoder.