Decoding human non-coding disease genetics en masse using Micro Capture-C and Deep Neural Network Machine learning.

Our proposal is an ambitious plan of multidisciplinary research combining cutting edge genomics and Machine Learning (ML) technologies to map the regulatory wiring of the human genome in unprecedented detail. Its outputs will provide immediate gains in decoding non-coding disease genetics, as demonstrated by our recent decoding of a locus associated with COVID-19 severity enriched in South Asian populations. We will leverage these unique maps to develop highly validated predictive ML methods to identify which variants are causal, which cells types they affect and importantly the identity of the genes and pathways that underlie the genetics of common and rare diseases. We will recursively train these networks using single-cell epigenetic data from previously inaccessible cell types, such as the human brain, to overcome the current barriers to decoding disease genetics in these tissues. Finally, we will thoroughly validate these approaches via worldwide collaboration as part of ICDA and within our own programme in haematological primary cell types solving many of the common variants associated with infection and autoimmunity. Our ultimate goal is to make sequence changes in the non-coding genome as interpretable as in the coding genome.