A pathway-informed disease-related gene identification approach and its application to screen novel risk genes for Alzheimer's disease.

[BACKGROUND] Alzheimer's disease (AD) is a severe neurodegenerative disorder, yet its molecular mechanisms remain incompletely understood. It is known that the joint action of a number of genetic and other factors is involved in the pathogenesis of this disorder.

[OBJECTIVE] In the past years, extensive research has focused on identifying novel AD predictors and genetic markers. However, our understanding of the molecular features of AD is still incomplete, and it is essential to discover novel genes and their interaction involved in the etiology and development of AD.

[METHODS] Here, we developed GRESA (Gene RElationship Sequence Analyzer), to predict novel genes related to disease by a combination of machine learning algorithm Skip-gram and Monte Carlo Tree Search (MCTS). First, we extracted gene association information contained in Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database and generated gene relationship sequences. Then, Skip-gram neural network was trained and combined with MCTS to predict genes potentially associated with diseases like AD.

[RESULTS] The performance of GRESA was evaluated on breast cancer and gastric cancer. We further constructed the gene network underlying AD pathogenesis via GRESA, by which 164 potential gene candidates were predicted and their molecular function and biological features were investigated.

[CONCLUSIONS] Results from this study provided insights for understanding the molecular feature underlying AD. As a useful systematic method, our approach can also be applied to construct gene networks for other complex diseases.