Identifying Blood-Based Biomarkers for Onset of Autism Spectrum Disorder (ASD)

Overview

ASD is a neurodevelopmental condition that affects up to 3% of children. Though behavioral tests exist, there are no definitive genetic tests that measure the onset of ASD, leaving 25% of individuals with symptoms undiagnosed. We aimed to solve this issue by identifying blood-based genes indicating ASD onset, licensing physicians to use minimally invasive diagnostic techniques, such as blood tests, to detect high risk genetic biomarkers for ASD. We created a heterogeneous graph made of patient and gene nodes connected by patient-gene edges derived from transcriptomic data and gene-gene edges built from STRING coexpression scores. We trained this data on a graph neural network (GNN) to predict ASD in patient nodes across 4-fold cross-validation. We trained a graph with brain and blood gene nodes and a graph with blood gene nodes containing differentially expressed gene (DEG) brain embeddings. The accuracy of the blood-only architecture with brain DEG gene embeddings was 62.5%, outperforming the brain and blood architecture GNN by 4.2% and the Elastic Net regression model by 40.6%. The model performed better when trained on blood data with prior brain expression bias, suggesting that the model predictions were based on neural pathology. Our gradient attribution analysis revealed a lack of alignment with established risk genes likely due to low expression levels in blood. These findings highlight genes enriched in growth factor signaling and immune-related tissue expression that provide motivation for further investigation and functional validation.

Competition history

• CSEF 2026 Medicine & Physiology (Senior Division) · Entry S-15-16

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Biomedical and Health Sciences projects
• CSEF Biomedical and Health Sciences projects
• Every CSEF 2026 project