MOGT Framework Identifies Brain Disorder Genes

Researchers have developed a computational framework designed to identify genes associated with brain disorders and support drug repurposing efforts. The Multi-omics Graph Transformer Network (MOGT) utilizes graph representation learning to model biological networks from multi-omics data, aiming to predict disease-associated genes more accurately than current methods [2]. Published in PLOS Computational Biology, the study highlights the framework's ability to link genetic findings with disease mechanisms for conditions like schizophrenia and Parkinson's disease [3].

Key takeaways

• MOGT is a semi-supervised graph neural network that uses multi-omics data to predict disease-associated genes [2].
• The framework outperformed existing approaches in identifying genes for two psychiatric disorders and three neurodegenerative diseases [2].
• Researchers identified 10 potential drug candidates for Parkinson's disease, with one drug showing positive results in experimental verification [2].
• The study focuses on complex conditions including schizophrenia, Alzheimer's disease, and Parkinson's disease [3].

Bridging genomics and deep learning

Genome-wide association studies (GWAS) have identified many genetic loci linked to neuropsychiatric and neurodegenerative disorders, but understanding how these loci impact disease remains unclear [2]. To address this, the authors proposed MOGT, a semi-supervised graph neural network that leverages deep learning to merge diverse genomic signals [3]. The framework models biological networks derived from multi-omics data to predict high-risk genes (HRGs) for brain disorders, outperforming current approaches in disease gene prediction for two psychiatric disorders and three neurodegenerative or neurological diseases [2]. The authors collected data from public databases such as the Human Phenotype Ontology (HPO), Online Mendelian Inheritance in Man (OMIM), and GeneCards to construct these networks [3].

From gene prediction to drug discovery

The researchers applied MOGT to facilitate drug discovery, specifically focusing on Parkinson's disease (PD). By integrating high-risk genes predicted by the network with the CMAP database, the study identified 10 drugs as potential candidates [2]. Among these, the effect of a drug named UK-356618 was experimentally verified in a primary neuron model. The results showed that this drug reversed the abnormal expression of PD-associated genes and improved cell-level phenotypes, suggesting the framework's utility in identifying novel treatments [2]. The study emphasizes that the predicted HRGs can be used to facilitate novel drug candidates, offering a pathway to translate complex genomic data into therapeutic options [3].

Why it matters

The authors suggest that MOGT provides high-level insights into the mechanisms and potential treatments of brain disorders [3]. The study notes that while previous computational models were often designed for cancers, this framework addresses the specific complexity of brain-related conditions where many small genetic changes play a role [3]. To support further research, the authors have made the source code available on GitHub, allowing other researchers to train the MOGT model and reproduce the results [3].