Classification of Autism Genes Using Network Science and Linear Genetic Programming

Understanding the genetic background of complex diseases and disorders plays an essential role in the promising precision medicine. Deciphering what genes are associated with a specific disease/disorder helps better diagnose and treat it, and may even prevent it if predicted accurately and acted on effectively at early stages. The evaluation of candidate disease-associated genes, however, requires time-consuming and expensive experiments given the large number of possibilities. Due to such challenges, computational methods have seen increasing applications in predicting gene-disease associations. Given the intertwined relationships of molecules in human cells, genes and their products can be considered to form a complex molecular interaction network. Such a network can be used to find candidate genes that share similar network properties with known disease-associated genes. In this research, we investigate autism spectrum disorders and propose a linear genetic programming algorithm for autism gene prediction using a human molecular interaction network and known autism-genes for training. We select an initial set of network properties as features and our LGP algorithm is able to find the most relevant features while evolving accurate predictive models. Our research demonstrates the powerful and flexible learning abilities of GP on tackling a significant biomedical problem, and is expected to inspire further exploration of wide GP applications.