bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2022, Номер unknown

Опубликована: Май 3, 2022

Abstract Multi-omic data spanning from genotype, gene expression to protein have been increasingly explored interpret findings genome wide association studies of Alzheimer’s disease (AD) and gain more insight the mechanism. However, each -omics type is usually examined individually functional interactions between genetic variations, genes proteins are only used after discovery findings, but not beforehand. In this case, multi-omic likely functionally related therefore give rise challenges in interpretation. To address problem, we propose a new interpretable deep neural network model MoFNet jointly prior knowledge set. It aims identify subnetwork predictive AD evidenced by measures. Particularly, interaction was embedded into architecture way that it resembles information flow DNA protein. The proposed significantly outperformed all other state-of-art classifiers when evaluated using ROS/MAP cohort. Instead individual markers, yielded sub-networks innate immune system, clearance misfolded proteins, neurotransmitter release respectively. Around 50% these were replicated another independent Our identified gene/proteins highly synaptic vesicle function. Altered regulation or genes/proteins could cause disruption neuron-neuron neuron-glia cross talk further lead neuronal synapse loss AD. Further investigation possibly help decipher mechanisms underlying dysfunction AD, ultimately inform therapeutic strategies modify progression early stage.

Язык: Английский