Bilkent University
Department of Computer Engineering


ASD-Tailored Biclustering Model for Early Autism Disorder Diagnosis


Alper Eroğlu
MS Student
Computer Engineering Department
Bilkent University

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental disorder and is the result of interplay between estimated one thousand genes. Despite sequencing close to five thousand ASD families, researchers have discovered only sixty-five ASD-risk genes. Thus, we have long ways to go for fully understanding the genetic architecture. We know only a small piece of a large puzzle and this makes it almost impossible to diagnose children by just considering their genetic profiles (mutations). To make matters worse, medical doctors cannot diagnose ASD phenotypically either, because of its comorbidity with other disorders at early ages (e.g., ADHD). In the absence of clear phenotypic cues, an only-genetic profile based classifier would enable early intervention and putting children in personal education tracks. As a testament to the difficulty of the problem, a recent study has shown that such a classifier can only have a true-positive rate of 0.08 when false discovery rate is restricted. In this project, we propose an ASD-tailored biclustering model to achieve significant automated classification results, which would enable a diagnosis much earlier than what is currently possible.


DATE: 06 May 2019, Monday, CS590 presentations begin at @ 15:40