A Semi-Supervised Generative Adversarial Network for Prediction of Genetic Disease Outcomes

• Caio Davi  |  Ulisses Braga-Neto

• Department of Electrical and Computer Engineering Texas A&M University College Station  |  USA

For most diseases, building large databases of labeled genetic data is an expensive and time-demanding task. genetic Generative Adversarial Networks (gGAN), a semi-supervised approach based on an innovative GAN architecture to create large synthetic genetic data sets starting with a small amount of labeled data and a large amount of unlabeled data is proposed. Even though dengue is being used to demonstrate the method, it is entirely general and can be applied to other diseases. The goal is to determine the propensity of a new individual to develop the severe form of the illness from their genetic profile alone. This model is self-aware and capable of determining whether a new genetic profile has enough compatibility with the data on which the network was trained and is thus suitable for prediction.

Input variables : 1. Unlabeled genetic data from Phase 3 of the 1000 Genomes Project which consists of 2504 individuals genotyped for more than 84 million variants 2. Labeled data - genotypes measured at 322 loci polymorphisms for the Dengue Infection phenotype in human subjects
Output Variables : 1.Unlabeled output - the genetic profile is real or synthetic 2. Labeled output - whether the individual with the corresponding genetic profile is likely to develop Severe Dengue

Visit Model : github.com

Additional links : arxiv.org