Conditional variational autoencoders for generated hub genes in human endogenous retrovirus and periodontal inflammasome

Autocodificadores variacionales condicionales para genes centrales generados en retrovirus endógenos humanos e inflamasoma periodontal

Abstract

Introduction: Human endogenous retroviruses (HERVs) represent about 8 % of the human genome, with HERV-K being implicated in various autoimmune conditions. Recent studies suggest its role in periodontal disease, a chronic inflammatory condition. Objective: This study aims to use Conditional Variational Autoencoders (CVAE) to generate hub genes involved in the interaction between HERV-K and periodontal inflammasomes, potentially uncovering novel therapeutic targets for periodontal disease. Methods: Two gene expression datasets from the NCBI Gene Expression Omnibus (GEO) (GSE240644 and GSE262663) were analyzed for differentially expressed genes (DEGs) using GeoRtool. DEGs were further examined using Cytoscape, incorporating the GeneMANIA plugin to construct an interactome of 500 genes. CytoHubba’s Maximum Centrality Clique method was employed to identify hub genes. The dataset was divided into hub and non-hub gene labels, then subjected to a CVAE model with 80 % training and 20 % testing. Model optimization involved hyperparameter tuning, enhancing hidden and latent dimensions, and adjusting the number of training epochs.