|
Abstract:
|
Modeling the link between neurocognition and brain physiology may fail to yield insightful inference without an accurate biological characterization of its relationship. Instead of directly modeling this relationship, for example, with a linear model between clinical traits and neuroimaging predictors, we consider pairwise similarity of neuroimaging features between subjects in a sample through the Similarity Model, which requires very minimal assumptions on underlying biology. From this Similarity Model, we propose a novel method to assess risk scores for neurocognitive deficits using neuroimaging data through a clustering-based algorithm. The interaction between these risk scores and genomic factors are then empirically examined to identify genomic markers having a significant impact on learning ability from the Pediatric Imaging, Neurocognition, and Genomics (PING) study. We observed there that the gene × risk score interactions improve power in a genome-wide association study. We then identify SNPs that achieving genome-wide significance for association with learning ability in samples from both the PING study and a replication study.
|
