Graph Node Classification using Autism Genetic Interaction Network

A Conversation with Kyle Riccardi

Autism Spectrum Disorder (ASD) is a heavily researched topic, but with very few answers. ASD networks show the interaction between genes that are both related and not related to ASD. The goal of this model is classified relationship each gene has to ASD. The model used for this dataset needed the ability to handle the lack of relationship towards ASD, so to solve this problem we used two techniques Up sampling and Graph Smote to fix our imbalance dataset. The dataset used is autism db network contains interaction between genes that do exist in ASD gene list and ones that don’t. The SFari dataset was used on top the autism db network to identify, which genes are related and not-related. The experiments analyzed with this dataset were Graph Convolutional Neural Network with an Up sample, Graph Sage with an Upsample, Graph Sage using Graph Smote, and Graph Convolutional Neural Network using Graph Smote. These modeled experiments were used a node feature mapping related to 1294 binary locations about the nodes. This feature mapping was used alongside the adjacency matrix of autsim db network to classify the relationship genes have with ASD, which showed that Graph Neural Networks could be used to classify genes related to ASD.

For more information about Fairfield check out:

+ http://www.fairfield.edu
+ http://www.fairfieldstags.com

Make sure to follow us on social media:

Instagram | http://www.instagram.com/fairfieldu
Twitter | http://www.twitter.com/fairfieldu
Facebook | http://www.facebook.com/FairfieldUniversity
Snapchat | fairfieldu


© Fairfield University 2022